Advances and Prospects in Fault Diagnosis of Aircraft Hydraulic Piston Pumps

Abstract

The hydraulic system is a crucial system for aircraft. As the core power component of the hydraulic system, the hydraulic pump has a complex structure and operates in a harsh environment. Therefore, fault diagnosis of the hydraulic pump is of great significance and highly challenging. This paper reviews the progress of fault diagnosis of aircraft hydraulic piston pumps. Firstly, it analyzes the failure mechanisms (such as wear of friction pairs, failure of central springs and bearings, cavitation of valve plates, etc.) and vibration mechanisms (fluid and mechanical vibrations). Then, it elaborates on traditional methods (model-based and signal processing-based), traditional data-driven methods (such as classifiers like support vector machines [SVM], extreme learning machines [ELM] combined with signal features), and machine-learning methods (single-signal, multisignal, and combined AI methods). Current research focuses on the improvement of signal processing and machine-learning classification, but there are still pain points in engineering applications. The main future trends are as follows: deepening the research on multisource signal fusion to improve diagnostic performance; overcoming the small-sample problem with the help of sample enhancement and transfer learning; and achieving automated online diagnosis using cloud computing.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

1. Introduction

The hydraulic system is a crucial component of aircraft, which is responsible for providing precise actuation control for structures such as flaps, speed brakes, and landing gear [1], and its operational stability is of paramount importance to flight safety. The piston pump, as shown in Figure 1, characterized by its compact structure, low flow rate, high pressure, and ease of variable control, is the most widely chosen option for high power density sources in aircraft hydraulic systems [2]. It is also the most core power component within the hydraulic system.

[figure(s) omitted; refer to PDF]

As a high-precision multibody moving component, the aviation piston pump boasts a complex internal structure and operates under harsh conditions, making it inaccessible for timely manual maintenance. During the initial stages of the malfunction, it is often challenging to promptly ascertain the cause and location of the issue. As the fault evolves and progresses, it may lead to the destruction of the pump’s internal structure, ultimately resulting in catastrophic consequences and posing significant risks to flight safety. In fault handling, fault diagnosis is the first and most critical step. It requires identifying the fault type and location to select the optimal repair method. Therefore, fault diagnosis technology is particularly important to guarantee the safety of the entire equipment and reduce the incidence of accidents. Currently, the mainstream approach for detecting the operational status and faults of piston pumps involves utilizing sensors to collect signals related to vibration, pressure, flow, and sound from the piston pump. By extracting characteristic information from these signals, the health status of the piston pump can be identified.

Due to the high levels of signal noise and the subtle characteristics of fault signals in piston pumps, signal-based fault detection methods continue to face numerous challenges. In response, extensive research had been conducted both domestically and internationally, employing a variety of methods to achieve fault diagnosis in piston pumps. These methods mostly encompass stages such as signal acquisition, noise reduction, feature extraction, classifier training, and diagnosis [3]. Among these methods, the construction of feature extraction and classifiers are the key elements. Consequently, fault diagnosis methods for piston pumps can generally be categorized into the following three types in recent years: those based on physical models, signal processing, and artificial intelligence (AI) techniques. Meanwhile, traditional experience-based methods such as fault tree and expert systems have received less attention. All methods have their own advantages and disadvantages, as shown in Table 1. Model-based methods contain physical information with strong diagnostic interpretability but are heavily dependent on experience; signal-based methods are simple but require manual feature extraction; and AI-based methods can automatically learn the fault features, but they possess the weakest interpretability and have certain requirements for the sample size of the fault signal [4]. With the development of research, the piston pump fault diagnosis methods have been greatly improved and novel hybrid methods have emerged. In order to clarify the development trends in fault diagnosis methods of aviation piston pumps, this paper summarizes the fault diagnosis research in recent years. It begins with a brief introduction about the fault mechanisms, then the traditional methods and AI methods are summarized, respectively, and finally the research status and development trend are summarized.

Table 1

The methods, core principles, and advantages of fault detection.

Methods	Core principles	Superiority
Traditional fault diagnosis methods	Developing fluid dynamics/mechanism motion models and extracting time/frequency/time-frequency domain features from vibration signals	Training-data-independentLow computational overheadHigh interpretability
Traditional data-driven methods	Performing classification/clustering on handcrafted features	High computational efficiencyMedium-scale data demandsPhysically traceable diagnostics
Artificial intelligence-based methods	Automatically learning features from raw data; generating fault data via high-fidelity models; training AI models	Automated deep feature extractionHigh-accuracy complex fault identificationFew-shot learning advancements

Current fault detection methodologies for aviation hydraulic pumps have evolved into a four-tiered technical architecture as follows: Signal Acquisition Layer: Multiphysical signals (e.g., vibration, pressure, and acoustic) are collected via heterogeneous sensor arrays. Feature Processing Layer: Fault-sensitive features are extracted through time-frequency analysis and modal decomposition techniques. Intelligent Diagnosis Layer: Fault pattern recognition is achieved via conventional classifiers or deep learning models. Decision Application Layer: Cloud computing platforms enable condition assessment and predictive maintenance.

2. Mechanism

2.1. Failure Mechanism

After a certain period of operation, the components of the piston pump will gradually experience wear and damage; fracture failures occur occasionally. When operating under abnormal working conditions, various fault phenomena may arise, such as increased noise, enhanced vibration, and reduced flow rate [3]. The hydraulic piston pump structure is complex, which contains a large number of relative-motion, accompanied by intense friction, such as a piston ball head and slipper, slipper and swashplate, piston and piston hole, and cylinder and valve plate. The wear of these friction pairs is the most common failure mode of the piston pump [5]. There are four typical faults, as shown in Table 2.

Table 2

Summary of four typical faults [5].

Fault type	Fault components	Fault cause
Loose slipper	Slipper	Dust in the working environment enters the gap between the piston ball head and slipper. Less precise dimensions of piston ball head and slipper
Slipper wear	Slipper	The supporting force of the swashplate on the slipper changes and makes the oil film thinner. The existence of a wedge-shaped clearance between the swashplate and the transmission shaft.
Center spring failure	Center spring	The long-term and high-intensity operation of the HPP causes fatigue wear and plastic deformation of the center spring.
Valve plate wear	Valve plate	The high-speed operation of an HPP increases the centrifugal force of the cylinder. Changes in the inclination angle of the swashplate change the thickness of the oil film.

The contact between the slipper and swashplate is not only the most stressed but also the most complex working environment in these friction pairs. The proper contact of the friction pair plays an important role in the normal operation of the machine while sealing the gap between the slipper and swashplate, but also providing a combined sliding support for the slipper [6]. Figure 2 shows the structure diagram of the slipper and plunger.

[figure(s) omitted; refer to PDF]

The slipper and swashplate cooperate in the form of floating noncontact static unloading. There is a small hole in the center of the end surface of the slipper, through which the high-pressure oil in the piston cavity can flow out, producing an oil film between the slipper and swashplate to avoid direct contact friction between the two metal parts. The huge impact forces when the hydraulic pump starts and stops or the abnormal vibration of the hydraulic pump, resulting in the rupture of the oil film between the slipper and swashplate, which will lead to the wear of the slipper [7]. Figure 3 shows the photos of the wear condition of the slipper and swashplate. In addition, if the oil is mixed with impurities, the wear will be further aggravated; when the degree of wear reaches a certain level, the slipper is likely to be seriously damaged.

[figure(s) omitted; refer to PDF]

For the purpose of wear resistance, the material of the slipper is usually made of brass or alloy powder. Therefore, in the operation of the piston pump, due to the significant periodic mutual impact forces between the piston ball head and the piston, it may cause plastic deformation of the slipper, leading to an increase in the clearance of the friction pair, which is known as “loose slipper.” When the slipper becomes loose to a certain extent, it may even detach. Research has also indicated that dust and other impurities in the friction interface can lead to loose slipper failure [5].

In addition to the above wear failures, center spring failure and bearing failure are also typical forms of failure. The center spring is assembled in the piston pump to provide preload to ensure the sealing effect between the slipper and the swashplate, as well as between the valve plate and the cylinder. During the long-term and high-intensity operation in the hydraulic piston pumps, the central spring may refer to the fatigue wear and plastic deformation due to the long-term pressure, resulting in stress relaxation, which reduces its preload, and the sealing effect is lost. The rolling bearing will gradually form cracks inside under the action of long-term alternating load. These cracks gradually develop to the surface with the continuous load and eventually cause the metal to form a pit, leading to bearing failure [8].

Aviation hydraulic pump failures primarily involve friction pair wear, cavitation erosion under high pressure, bearing failures, and structural fatigue. Damage interaction emerges when these mechanisms coexist, creating compounded degradation pathways.

2.2. Vibration Mechanism

The entire vibration of the plunger pump is generally separated into two sections: fluid vibration and mechanical vibration [9].

The main part of the fluid vibration is the periodic pulsation. The piston pump performs oil absorption and discharge by changing the volume in the piston cavity, which relies on the reciprocating movement of the piston; thus, the inhomogeneity of instantaneous flow produces periodic fluid pulsation. The frequency of the vibration is related to the working frequency of the pump and the number of pistons.

During the discharge stage, when the low-pressure oil in the cavity is connected with the high pressure oil in the oil drain area, the high-pressure oil will quickly flow into the cavity because of the pressure difference, forming a backflow to cause vibration. Cavitation occurs when localized pressure in piston pump suction chambers drops below either the dissolved-air release threshold or the fluid’s vaporization point. Hydraulic oil often contains a certain amount of dissolved gas, which both lead to the cavitation phenomenon inside the pump. The collapse of bubbles in the high-pressure zone of the distribution plate can also cause vibrations [10]. The vibration signals of the plunger pump were collected by vibration sensors and input into the acquisition card. Subsequently, the virtual acquisition program in LabVIEW software was utilized for the acquisition and visualization of multiple signals, with the data ultimately stored in the computer [11]. Increasing the oil suction pressure can suppress the occurrence of cavitation phenomena [12].

The mechanical vibrations mainly include the following: (1) vibrations caused by the swashplate and variable mechanism under the action of periodic torque; (2) vibrations resulting from defects in the bearings themselves and improper installation procedures; and (3) vibrations arising from eccentricity or imbalance of the pump’s rotating components during operation [13].

It is evident that when the operating status changes, it inevitably leads to alterations in the motion status of the internal components of the pump, ultimately manifesting as changes in the overall vibration condition of the piston pump. For instance, the rotational speed determines the fundamental frequency and its harmonics, with most harmonic vibration responses increasing as the discharge pressure and displacement angle rise [14]. Moreover, the endpoints of vibration state changes caused by different fault modes are not the same. Therefore, there is a natural rationality to detect and diagnose piston pump faults through vibration signal acquisition, feature extraction, and classification.

Furthermore, compared to other signals such as temperature, pressure, and flow rate, the vibration signals exhibit discontinuity and nonlinearity, often containing more fault-related information. At the same time, the measurement of vibration signals is relatively straightforward, and they are more significantly influenced by the operating state. Therefore, among the fault diagnosis methods for hydraulic pumps based on single signals, the use of vibration signals for diagnosing hydraulic pump faults is currently the preferred approach in most studies.

The precise measurements of hydraulic pump vibration signals require consideration of their vibration mechanism characteristics. For fluid vibration measurement, a waterproof accelerometer is adopted, or sensors are arranged within 5 cm of the high-pressure oil port to capture fluid pulsation. For mechanical vibration measurement, a triaxial accelerometer is rigidly mounted on the bearing housing to enable synchronous acquisition with rotational speed information. Fluid and mechanical vibration components are distinguished by short-time energy analysis, and the signal validity is ensured through a coherence function test.

3. Traditional Fault Diagnosis Methods

3.1. Model-Based Methods

Model-based methods establish numerical or analytical models to reveal system failure evolution mechanisms and construct mapping relationships between fault modes and signal responses. Their effectiveness is primarily reflected in three aspects: Firstly, by comparing actual equipment operating parameters with model simulation results, rapid fault detection and precise identification can be achieved. Secondly, this approach significantly reduces reliance on experimental data and improves diagnostic efficiency. Lastly, the model-based diagnostic framework provides theoretical foundations for fault mechanism analysis and predictive maintenance [15]. Due to highly coupled internal structures and diverse fault modes, the model’s accuracy and generalization capability still face challenges, which to some extent limit the engineering applicability of this diagnostic method. Future research should focus on enhancing the model’s characterization ability of complex system dynamics to further improve its practical value.

Ma et al. [16] established a simulation model of an aircraft hydraulic system based on the AMEsim platform. They obtained real parameters for fault monitoring by analyzing the relationship between flight parameters and hydraulic system fault modes. Bensaad et al. [17] developed a nonlinear dynamic model for leakage faults in plunger pumps and implemented this model on the Simulink platform. They then used an extended Kalman filter to estimate the pressure within the plunger cavity and validated the model through measured pressure data, achieving the identification of worn plungers. Shi [18] created a mathematical model linking the wear clearance of friction pairs to leakage flow and validated the theoretical model through internal flow field simulations. The variational mode decomposition (VMD) method was applied to process pressure and flow signals to assess the degradation state of the plunger pump’s friction pairs.

Yang et al. [19] modeled a swashplate-type piston pump based on the Modelica, as shown in Figure 4. They clarified the impact of different fault modes on the hydraulic system through fault injection and quantitative simulation analysis.

[figure(s) omitted; refer to PDF]

Chen Shuai [20] established a rigid-flexible-fluid coupling model with piston ball head crack fault by ADAMS and AMEsim. By performing co-simulation, simulated vibration signal samples were obtained. Experimental samples were used to train a probabilistic neural network (PNN), which was then employed to diagnose faults in the simulated samples, effectively verifying the reliability of the simulation model.

Guo et al. [21] introduced a model-based Stochastic Differential Mode Decomposition (SDMD) method; it is the finite element method that determines the central frequency of a faulty plunger pump to construct an optimal filter for extracting the fault resonance frequency band. Dai [22] established a rigid-flexible coupled hydraulic model based on ANSYS APDL to analyze the vibration characteristics of faults.

To target the fatigue damage of plungers under complex alternating stress, Tang et al. [23] established a rigid-flexible-liquid coupling model. They investigated the relationship between indicators such as rotational speed, pressure, and the remaining lifespan of the plunger by combining the finite element method with Miner’s linear damage accumulation rule, and then they constructed a multidomain coupling model of an axial piston pump based on the simscape platform [24]. By integrating fault injection technology and MATLAB’s fast restart function, they obtained pressure and flow data to identify faults successfully.

3.2. Signal-Based Methods

After denoising the original signals, techniques such as Fourier transform (FT), principal component analysis (PCA), wavelet transform (WT), empirical mode decomposition (EMD), and VMD, along with their variants, are employed to transform, decompose, or reconstruct the signals. This process extracts fault signal features, which are used to detect and classify faults in the piston pump.

3.2.1. Frequency-Domain Analysis

Przystupa et al. [25] discovered that short-time FT (STFT) could clearly capture flow disturbances to identify the operating conditions better than Fast FT (FFT). Hu et al. [26] proposed a fault diagnosis method based on multifeature threshold criterion fusion for typical piston pump faults. They diagnosed faults under complex working conditions by performing threshold judgments on multisource signals such as vibration, pressure, and flow. Qi et al. [27] extracted features from the frequency-domain signals of vibration sensors on piston pumps and established a fault-model database to enable online fault diagnosis. They also considered the contribution rates of different sensors, reducing computational load while effectively improving diagnostic accuracy. Xia et al. [28] constructed a mathematical model of discharge pressure for piston pumps with cavitation damage and conducted experimental studies on it. They found that cavitation introduces new frequencies and sidebands in the discharge pressure frequency domain, which could detect specific cavitation damage in axial piston pumps.

3.2.2. Spectrum

Zheng et al. [29] proposed a feature extraction method using power spectral entropy (PSE) as a substitute for kurtosis in autocorrelation graphs. As shown in Figure 5, they validated the method using the Case Western Reserve University bearing dataset, suppressing noise while highlighting frequency-domain information features, it could identify slipper wear faults based on vibration signals and outperformed traditional autocorrelation graph methods in diagnostic performance. The flowchart of PSE-Autogram is displayed in Figure 6.

[figure(s) omitted; refer to PDF]

Hemati et al. [30] studied a case of gear failure caused by bearing loosening, analyzing the vibration signals using envelope spectrum and acceleration spectrum methods. They found that the method effectively captured gear meshing waves and rotational speed harmonics excited by bearing loosening and gear tooth wear. Zhou et al. [31] established a connection between cavitation and outlet pressure fluctuations. They discovered that the sideband energy integral had high discriminative power for cavitation states. By collecting outlet pressure signals from the piston pump and calculating the sideband energy integral of harmonic components in the low-frequency range, they distinguished different cavitation states of the piston pump. Kumar et al. [32] proposed a noise subtraction and marginally enhanced square envelope spectrum method, which uses vibration signals from normal piston pumps to denoise fault vibration signals. This approach significantly reduces deterministic vibration components in fault signals, thereby highlighting subtle abnormal features caused by faults.

3.2.3. Wavelet

Yu et al. [33] improved the empirical WT (EWT) and introduced a novel signal fusion framework, which can highlight fault frequencies in noisy by dynamically fusing three-channel vibration signals using variance contribution rates with high-accuracy at low computational cost. To address the issue of overdecompose in the EWT method; Zheng et al. [34] applied an improved EWT method, which replaces the Fourier magnitude spectrum with the power spectrum for segmentation. Qiao et al. [35] determined the frequency-sensitive range in the wavelet packet energy spectrum based on the vibration mechanism, the fault characteristics could be highlighted effectively.

3.2.4. Mode Decomposition

Wang et al. [36] extracted a fault feature method to reconstruct effective signals, combined Fast EMD with relative entropy. Aiming to highlight feature signals, Deng [37] utilized the extreme-point symmetric mode decomposition-Teager, causing characteristic peaks to appear in the spectrum and ultimately obtaining effective feature vectors. To extract periodic pulses of fault signals in noisy environments, Xiao et al. [38] applied the Multipoint Optimal Minimum Entropy Deconvolution Adjusted (MOMEDA) to deconvolve vibration signals. They optimized the MOMEDA filter using the advance-retreat method and then employed the Teager energy operator for demodulation to obtain frequencies, thereby achieving fault diagnosis for axial piston pumps.

3.2.5. Other Methods

Zheng et al. [39] proposed a novel fault diagnosis method based on complex signals. By collecting vibration signals from two orthogonal sensors on the piston pump and fusing them into a complex signal, they performed envelope analysis using the log-Spectrum Amplitude Modulation (log-SAM) method to achieve fault diagnosis. Li et al. [40] introduced a degradation feature extraction method for hydraulic pumps based on relative entropy. By following the maximum correlation entropy criterion and a step-by-step fusion strategy, they fused initial data into effective features. Liu et al. [41] proposed a data augmentation method based on Symplectic Geometric Mode Decomposition (SGMD). They expanded the sample size while ensuring high similarity to fault samples by randomly selecting, enhancing, and reconstructing SGMD components. Casoli et al. [42] acquired vibration signals through time sampling and applied PCA to reduce the dimensionality of feature vectors obtained from the FFT of the vibration signals. Tang et al. [43] addressed the issue of diagnosing loose slipper faults in variable-load piston pumps, discovering that the gradient of the axial vibration root mean square trend line could sensitively reflect loose slipper faults and was unaffected by external load variations. Buiges et al. [44] performed PCA on multisource signals such as acceleration, flow, and pressure to preliminarily classify the blockage state of hydraulic pumps. Gu et al. [45] conducted research on detecting piston pump faults using instantaneous speed fluctuation signals, leveraging their noise resistance and angular domain stationarity characteristics.

Model-based methods require the establishment of precise mathematical models based on equipment failure mechanisms. However, the fuel control system of an aeroengine is highly complex, involving multidomain coupling (mechanical, electrical, and hydraulic), making it challenging to develop high-fidelity system models. Signal-based methods, while independent of physical models, still rely heavily on expert knowledge and exhibit limited adaptability. With the rapid advancement of China’s aeroengine industry, traditional fault diagnosis approaches can no longer meet the growing demands for high reliability and intelligent diagnostics.

4. Traditional Data-Driven Methods

After extracting fault signal features through signal processing and other methods, fault detection and diagnosis can be performed. However, in practice, avoiding manual classification, this approach is often combined with classifiers to achieve intelligent diagnosis. Traditional classifiers typically include data-driven classification methods such as SVM, extreme learning machines (ELM), Random Forest, and K-Nearest Neighbors (K-NN).

4.1. SVM

Zhao et al. [46] proposed a method that combines local mean decomposition with SVM. Compared with the diagnosis result of the original signal, diagnostic accuracy of the proposed method can reach above 99.5%. Chao et al. [47] introduced a density-weighted support vector data description (SVDD) method, which trains the density-weighted SVDD using only healthy piston pump signal samples. Experiments showed that this model can assess the health status of axial piston pumps for overcoming the challenges of collecting fault samples in practical applications.

4.2. ELM

The ELM is a classification algorithm based on Single-hidden Layer Feedforward Neural Networks, which exhibits better generalization performance than traditional learning algorithms. Moreover, ELM is less sensitive to parameters, enabling faster and more convenient classification and recognition of datasets, improving the traceability of damage to some extent.

For the wear condition of the slipper in piston pumps, Wenyan et al. [48] proposed a method combining Local S-Transform with ELM to identify wear conditions. Using only three feature vectors for classification, the method achieved a recognition accuracy of 99%. In subsequent research [49], to address the issue of weak vibration signal characteristics in early fault stages, they proposed a VMD based on Feature Energy Proportion (VMDF) method. The peak values of MDE of the reconstructed signals were used as feature vectors and input into the ELM for classification, enabling the identification of fault modes in axial piston pumps and reflecting the evolution of wear conditions. Jiang et al. [50] introduced a method based on Mel-Frequency Cepstral Coefficients (MFCC) combined with ELM. The acoustic signals were denoised using wavelet packet methods, and MFCC acoustic features were extracted and input into the ELM for training. This method accurately identified multiple fault modes and demonstrated that ELM has shorter training and diagnosis times compared to BP neural networks and SVM, showing promising application prospects. Li et al. [51] proposed an integrated fault diagnosis method based on modified ensemble EMD, autoregressive spectral energy, and wavelet kernel ELM. The results showed that the diagnosis accuracy rate can reach 100%, and the fault diagnosis time was only 0.002 s. Ding et al. [52] used PCA to reduce the dimensionality of feature vectors decomposed by EWT, and then input the key reduced-dimensional features into the ELM for learning. The method could process not only nonstationary signals but also stationary ones, making it generalizable and applicable to other pump vibration signals. Figure 7 illustrates the process of the proposed method.

[figure(s) omitted; refer to PDF]

4.3. Random Forest

Lee et al. [53] conducted research on the vibration signals of gear pumps with hydraulic oil contaminated by impurities. They implemented fault diagnosis based on vibration signals using Random Forest, K-NN, and AdaBoost algorithms. It is found that the Random Forest method achieved the highest accuracy but at a higher computational cost. Zhang et al. [54] investigated the characteristics of fault signals from piston pump slipper wear and the parameters of the Random Forest algorithm. Addressing the issue of fault data for axial piston pumps being significantly less abundant than normal operating condition data, Jiang and Ma et al. [55, 56] proposed a balanced random forest algorithm by combining undersampling with the traditional Random Forest algorithm. Testing on UCI datasets demonstrated that this method effectively improved the accuracy of fault classification under imbalanced data conditions.

4.4. Small-Sample

In order to overcome the difficulty of poor information of faulty piston pump, Jia et al. [57] applied Symbolic Peak-Interval-Pattern (SPIP) to the field of fault diagnosis for the first time, and proposed a data-driven fault diagnosis method combining SPIP and Hidden Markov Model (HMM). During the model training, the obtained PIP sequences are first symbolized, and a genetic algorithm is employed to optimally partition the symbolic space. These symbolized sequences are then fed into a HMM for recognition. In scenarios with limited sample points and short signal durations, this method achieves an impressive fault classification accuracy of up to 99.625%.

4.5. Others

Jiang et al. [58] extracted features based on recurrence quantification analysis and achieved fault recognition by combining the Kernel Fuzzy C-Means Clustering method. In subsequent work [59], they proposed a symmetric polar coordinate-based method to transform denoised vibration signals into snowflake images for fault type identification. The snowflake images were then converted into gray-level co-occurrence matrices, and fault diagnosis was performed using the FCMC. Jin et al. [60] implemented visual clustering and judgment for fault diagnosis of plunger pumps by combining the dynamic mode decomposition algorithm with t-distributed Stochastic Neighbor Embedding (t-SNE). Wang et al. [61] integrated VMD with Extreme Gradient Boosting Trees to classify different cavitation levels of piston pumps under noisy conditions. Fu [62] investigated the relationship between Bayesian network algorithms and hydraulic pump fault components based on simulation data. Under conditions of incomplete data, the Bayesian network algorithm was applied to perform fault diagnosis.

The traditional data-driven methods can realize the automatic diagnosis of the plunger pump fault, but they still face many limitations, with the primary bottleneck being their reliance on manually designed features. When dealing with multimodal data generated by complex equipment, these algorithms struggle to efficiently integrate different types of features. Moreover, traditional algorithms often cannot directly handle dynamic operating conditions or nonstationary signals, which limit their adaptability in practical applications.

5. AI-Based Methods

The AI-based method can learn the characteristics of the fault signal independently and realize the intelligent fault diagnosis. Therefore, the method has received the most widespread attention at present. There are many studies on it; various network models are fully explored. In this paper, the research status is classified according to signal types and learning models.

5.1. Single Signal

Among the AI-based fault diagnosis methods of a single signal, the vibration signal is also the most widely used.

Du et al. [63] built upon traditional neural network fault diagnosis methods by performing sensitivity analysis on vibration signals reconstructed using EMD. They diagnosed the fault by extracting highly sensitive feature parameters as inputs to train a PNN. The experiment showed that SA-PNN could quickly and accurately diagnose the fault of the plunger pump. However, the SA-EMD-PNN had higher diagnosis accuracy, and the method was promising for practical applications. Wei et al. [64] transformed fused vibration signals into frequency spectra and input them into a two-dimensional convolutional network for training, with which different cavitation levels were identified successfully. Wu et al. [65] addressed the issue of variable speed conditions by proposing a diagnostic method based on polynomial chirplet transform and VMD. They estimated instantaneous frequency using chirplet transform and performed resampling to convert vibration signals into angular domain stationary signals. Subsequently, they reconstructed the components obtained from VMD and conducted envelope order spectrum analysis. In order to consider sensor positions. Teng et al. [66] constructed a novel bagging-ensemble CNN fault diagnosis model incorporating multichannel sensor location information, with an accuracy rate of 92% on the test set and an average recall rate of 89% for various faults, overcoming the fault identification difficulties arising from vibration signal nonstationarity successfully. Chen et al. [67] achieved fault classification by employing MDE to fuse multichannel vibration signals and combined it with a Deep Residual Shrinkage Network. Fan et al. [68] constructed a fault diagnosis model for piston pumps based on a prototype network. The experiment results showed that the accuracy is more than 85% under the condition of limited samples while possessing certain noise immunity and Hu [69] developed the model using Graph Convolutional Network II. They generated feature graphs based on vibration signal feature vectors using the K-NN algorithm, and then trained the GCNII to achieve accurate fault classification. However, the experimental data was not collected from actual field operating conditions and remains at the laboratory application stage.

5.1.1. Small-Sample

One of the major disadvantages of AI-based methods is that the diagnostic accuracy rate depends heavily on the number of samples, while in practice, the sample signal of the faulty plunger pump is rare, which causes the imbalance of faulty/healthy samples, and brings some difficulties to the training of the model. So, small sample and sample imbalance has become one of the research hotspots at home and abroad nowadays.

5.1.2. Small-Sample Expand

An intuitive solution to the problem of scarce fault samples is to expand the existing fault samples to enhance the training dataset. Meng et al. [70] proposed Empirical Mode Reconstruction (EMR) to solve the issue of sample imbalance. The augmented samples exhibited highly similar features and categories to the real samples, enabling them to effectively guide the training of deep learning models. A series of experiments validated the effectiveness of the developed EMR method in diagnosing imbalance faults in civil aviation hydraulic pumps. Gao et al. [71] expanded the training samples using a Siamese neural network combined with a similarity comparison method to improve the accuracy under small sample conditions. With a small sample condition of 140 training samples, the recognition accuracy of the proposed neural network was more than 90%, which was 27.24% higher than 1D-CNN and 39.72% higher than SVM.

5.1.3. Small-Sample Transfer

An approach based on the transfer-learning model. Zhang et al. [72] proposed an adversarial domain adaptation approach simulated data-driven. By injecting faults into the plunger pump dynamics model, they simulated the vibration response in different modes, and then used the domain counter adaptive algorithm to learn the fault information from the simulated signal, so as to identify the faults in the measured signal.

5.1.4. Online

Aiming to enhance the automation of condition monitoring, researchers have explored online detection methods based on cloud platforms. Ma [73] proposed convolutional neural network (CNN) with wide first-layer Kernel, which can automatically learn fault oriented features, eliminating the time-consuming and laborious feature extraction process of traditional fault diagnosis, the method could be able to detect single-class faults in real time based on a cloud platform. Wang et al. [74] utilized a feature selection-based artificial neural network method, reducing delay by establishing nodes at the edge, and shortened the single fault diagnosis time for axial piston pumps to 0.24 s; it provided a technical route for online real-time fault diagnosis. Subsequently [75], they proposed a knowledge distillation model based on visual transformation to tackle the wear condition of key friction pairs in plunger pumps, achieving online recognition of the state of friction pairs.

Besides vibration signal features, fault diagnosis methods based on sound, pressure, flow, and other signals have been widely studied.

5.1.5. Acoustic Signal

Zhu et al. [76] employed the LeNet model to classify and learn acoustic signals, optimizing the parameters of the LeNet model using the particle swarm optimization algorithm to diagnose five different faults in plunger pumps accurately. Wang et al. [77] constructed multiple noncontact acoustic arrays to collect noise signals from piston pumps and trained a CNN-SVM combined model using multisignal fused time-frequency diagrams, enabling fault classification based on acoustic signals. The results showed that the classification accuracy of the proposed method reached 97.5% in the running noise environment, which was 1.1% higher than that of the single-channel time-frequency sample. Zhang [78] used a mean spectrum bar chart as a visual representation of plunger pump acoustic fingerprints and conducted research on the diagnostic accuracy of various network models, proposing a ResNet + self-attention transfer learning model based on domain adversarial training, as shown in Figure 8.

[figure(s) omitted; refer to PDF]

Similarly, Liu et al. [79] applied meta-transfer learning to ascertain features from single audio signals for achieving adaptive processing of unknown faults. Experimental results showed that the accuracy of the MTL-PAFD method can reach 91.41% for diagnosis of known fault classes. After fast adaptive learning, this method could achieve an accuracy of 89.64% when identifying unknown fault classes. Li et al. [80] proposed an adaptive noise reduction method for acoustic signals, where the signals were first denoised adaptively after Gammatone cepstral transformation and then input into a residual network for training and classification; thus, the fault diagnosis accuracy was improved to 96.97% after adaptive noise reduction.

5.1.6. Pressure

Zi et al. [81] proposed an iterative fault signal feature extraction method based on parametric demodulation. By parametrically demodulating the outlet pressure signals of piston pumps and reconstructing them into sample sets, they trained a CNN with short-term memory, ultimately achieving classification of piston pumps under different cavitation levels. Wang et al. [82] introduced a physics-informed neural network that transformed a physically integrated discharge pressure model into loss functions guiding the learning process, providing physical interpretations for parameter identification. This model identified wear faults in piston pumps based on the output equivalent clearance and volumetric efficiency, demonstrating robust diagnostic capabilities under multifault and high-pressure load conditions.

5.2. Multisource Signal

Benefiting from the automatic fault feature learning capabilities of AI methods, the complexity of multisource signal fusion diagnostic methods has been significantly reduced. Multisource signal diagnostic approaches can integrate signals of different forms, encompassing more comprehensive and higher-dimensional fault information, so as to enhance diagnostic accuracy.

5.2.1. CNN

Tang et al. [83–85] constructed a Batch Normalization CNN with a unified normalization strategy. They adaptively optimized the hyperparameters of the diagnostic model using a Gaussian process-based Bayesian optimization method and subsequently achieved stable fault diagnosis for hydraulic piston pumps by applying this model with synchronous compression wavelet transform. Comparative experiments demonstrated that the classification performance of this model showed significant improvement compared to the manually optimized LeNet-5 network. Zhu et al. [86] integrated a modified AlexNet construction with continuous wavelet transform for hydraulic piston pump fault diagnosis by reducing layer parameter quantities and computational complexity. By comparing the proposed model with standard LeNet-5, standard AlexNet, and 2D LeNet-5 networks, they found that their model achieved the highest diagnostic accuracy. Zhu et al. [87] applied the model to multisource signal S-transform images for fault identification based on BNCNN and introduced Bayesian networks to optimize model hyperparameters, the results showed that the proposed model achieved an average accuracy improvement of up to 11.57%. Wang et al. [88] utilized a multilayer CNN as the VAE’s encoder network, introducing a hybrid attention mechanism and adaptive Soft Thresholding (ST) to reduce noise and enhance feature extraction, as shown in Figures 9 and 10. The experiments revealed that 99.32% diagnosis accuracy under 5 dB noises and 69.72% under-5 dB noise, which outperforms commonly, used diagnosis methods.

[figure(s) omitted; refer to PDF]

Jiang et al. [89] developed a piston pump fault diagnosis algorithm based on EWT and 1D-CNN, targeting vibration and pressure signals. By deploying the model on a cloud platform, they realized real-time fault diagnosis for piston pumps via the Industrial Internet of Things (IIoT) cloud platform.

5.2.2. Fusion

Cui et al. [90] proposed a multisource signal diagnostic method based on particle swarm optimization-backpropagation and D-S evidence theory. After obtaining diagnostic results from sub-networks of different signal sources, they fused the results at the decision level using D-S evidence theory.

Zhu et al. [91] constructed a conflict coefficient between evidence based on a gravitational concept and developed a multisource signal fusion diagnostic model integrating BP neural networks and D-S evidence theory. The support degree of fusion diagnosis results was close to 1. Chao et al. [92] introduced a weighting scheme for multichannel vibration data fusion. The CNN received three channels of vibration data and made the final diagnosis through information fusion at the decision level. The experimental results showed that the proposed method improves the classification accuracy by adjusting the probability distribution of classification according to the learned weight matrix.

5.2.3. Other Networks

To address the issue of high-dimensional data in hydraulic systems, Li [93] proposed a fault diagnosis method combining Kernel PCA (KPCA) and PNN, which outperformed the conventional PCA-PNN method in both speed and accuracy. Liang et al. [94] developed the method based on a Siamese random spatiotemporal graph convolutional network. By leveraging random spatiotemporal graphs to capture spatial-temporal relationships in multisensor signals and integrating an enhanced Siamese neural network for learning and classification, their study demonstrated that the model maximizes the utilization of small-sample data for fault feature mining. Liu et al. [95] constructed a fault diagnosis framework using a temporal-spatial attention network. This method sequentially extracts fault-sensitive features from multisource sensor signals in the temporal and spatial domains, significantly improving diagnostic accuracy under coupled fault conditions.

Notably, ablation experiments revealed that the extraction order of temporal and spatial features critically impacts the model’s diagnostic capability. Dong et al. [96] designed a multiscale attention residual feature extraction network for domain-adversarial neural networks, enhancing the ability to extract multiscale features. Xu et al. [97] constructed a multiscale attention mechanism residual network (MSARN) to strengthen multisource information fusion and feature extraction of the variable-displacement pumps. The schematic diagram of dataset operating conditions and fault classification is illustrated in Figure 11. The 3-channel vibration signals are captured under thirteen operating conditions with four loose slipper degrees in the experiments.

[figure(s) omitted; refer to PDF]

The flowchart of the loose slipper fault diagnosis method based on the MSARN is illustrated in Figure 12, which comprised three primary steps: dataset construction, model training (including hyperparameter optimization), and model testing. By training the model under partially stable operating conditions, they achieved accurate diagnosis in time-varying operational scenarios.

[figure(s) omitted; refer to PDF]

5.3. Combined With AI

Given the abundance of machine learning models with varying strengths and limitations, novel composite AI methods have emerged as a research hotspot to fully leverage their performance advantages.

For axial piston pumps with diverse operating conditions and heterogeneous data distributions, deep learning models generally lack broad applicability. To address this, He et al. [98] proposed a multisignal fusion adversarial network based on deep transfer learning. This method incorporates a dynamic weight adjustment mechanism for vibration and acoustic signals and designs transfer scenarios based on workload. Demonstrating exceptional performance in cross-domain fault detection, it achieved an average accuracy of 98.5% across all scenarios.

Miao et al. [99] applied the TrAdaBoost transfer learning algorithm to train collected signal samples, iteratively adjusting sample weights based on their value. Comparative testing revealed that transfer learning significantly outperforms traditional machine learning methods in small-sample conditions and exhibits substantial advantages in variable condition fault diagnosis scenarios. Qiu et al. [100] integrated BP neural networks with the AdaBoost algorithm to construct a piston pump fault diagnosis model, showing marked improvement over conventional BP networks. Zhang et al. [101] combined the K-NN algorithm with the TrAdaBoost transfer model, using it to derive initial training set weights to enhance the training efficiency of TrAdaBoost.

Li et al. [102] proposed a CBLRE model for cavitation detection in piston pumps, fusing neural networks with classifiers. This approach employs CNN for signal feature extraction, bidirectional long short-term memory (BiLSTM) networks to handle temporal dependencies, and regularized ELM for fault classification, enabling comprehensive cavitation state monitoring and fault diagnosis.

AI-based methods hold revolutionary potential for fault diagnosis in hydraulic pumps. They demonstrate significant advantages, particularly in handling complex signals, mining deep-level features, and addressing the challenge of small-sample data. Current research hotspots and mainstream approaches center on deep learning and machine learning as the core technologies, combined with small-sample techniques like transfer learning and generative adversarial networks, and further integrated with physical model knowledge.

Despite facing severe challenges in data acquisition, model generalization, real-time deployment, and interpretability, AI-based methods are progressively making the transition from laboratory research to engineering applications. It stands as a critical technological enabler for building future high-reliability, intelligent aerospace hydraulic systems.

6. Conclusions

As a core component of aircraft hydraulic systems, piston pumps are critical to flight safety, and fault detection methods have been extensively studied since their initial application. In recent years, knowledge-based traditional approaches such as expert systems, fault trees, and signed directed graphs have fallen out of favor. Researchers worldwide now focus predominantly on fault signal processing techniques and machine learning-based classification methods, particularly exploring advancements in signal decomposition-reconstruction methods and machine learning model optimization.

Model-based methods are suitable for the simulation and prediction of specific faults but heavily rely on empirical modeling, resulting in limited generalization capability in practical engineering applications. Signal processing-based approaches, on the other hand, offer operational simplicity by directly extracting features from signals such as vibration and pressure, making them effective for fault detection in noisy environments. Traditional data-driven methods achieve high accuracy in diagnosing specific fault types; however, their performance is contingent upon the integration with signal processing techniques, and the classification efficacy is highly sensitive to feature quality. Machine learning methods minimize manual intervention, address small-sample challenges, and excel in handling complex nonlinear signals (vibration), positioning them as a pivotal direction for future research. However, translating advanced fault diagnosis methods into engineering practice requires addressing several research challenges. Future trends are anticipated to evolve in three key directions as follows:

1. Multisource Signal Fusion Optimization.

  Compare to single-source signals, multisource signal contains richer, higher-dimensional fault feature information, the fusion strategy significantly impacts diagnostic model performance. Further research on fusion methodologies remains imperative.

2. Small-Sample Learning Solutions

  Engineering scenarios often suffer from scarce fault samples and imbalanced health-fault data distributions, yet AI-based diagnostic accuracy heavily relies on training sample quality, data augmentation and transfer learning models are current solutions. Transfer learning frameworks trained on simulation-generated samples are emerging as a promising approach to overcoming small-sample limitations.

3. Automated Condition Monitoring and Online Diagnosis

  Existing fault diagnosis models typically require massive datasets and intensive computations. Leveraging cloud computing platforms’ computational power and storage resources enables online remote fault detection for hydraulic piston pumps, representing a pivotal direction for intelligent diagnostic systems.

Disclosure

All authors have read and agreed to the published version of the manuscript.

Author Contributions

Conceptualization, Peng Lin and Weijun Wang; methodology, Peng Lin; Weijun Wang; validation, Peng Lin, Weijun Wang and Yuting Xiong; formal analysis, Peng Lin; investigation, Yuting Xiong; resources, Peng Lin; data curation, Weijun Wang; writing–original draft preparation, Peng Lin; writing–review and editing, Yuting Xiong; visualization, Weijun Wang; supervision, Jingman Lu; project administration, Shu Wang; funding acquisition, Dong Hu.

Funding

This study was supported by the Science and Technology Program of Hunan Province, 2022RC1140; Natural Science Foundation of Hunan Province, 2023JJ50088; Hunan Provincial Department of Education Project, 22A0607; R&D program of Hunan Province, 2024AQ2001; and Aeronautical Science Foundation of China, 2023M026151001.

References

[1] Y. H. Li, X. Z. Wang, "Fault Diagnosis of Aircraft Hydraulic System," Computer Engineering and Applications, vol. 55 no. 05, pp. 232-236, DOI: 10.3778/j.issn.1002-8331.1711-0029, 2019.

[2] X. P. Ouyang, T. Z. Wang, X. Fang, "Research Status of the High Speed Aircraft Piston Pump," Chinese Hydraulics & Pneumatics, vol. no. 02,DOI: 10.11832/j.issn.1000-4858.2018.02.001, 2018.

[3] Y. F. Yang, L. Ding, J. H. Xiao, G. N. Fang, J. Li, "Current Status and Applications for Hydraulic Pump Fault Diagnosis: A Review," Sensors, vol. 22 no. 24,DOI: 10.3390/s22249714, 2022.

[4] R. Yan, W. G. Xu, Z. Y. Wang, "Research Status and Challenges on Fault Diagnosis Methodology for Fuel Control System of Aero-Engine," Journal of Mechanical Engineering, vol. 60 no. 04, 2024.

[5] Y. Zhu, H. Su, S. Tang, S. D. Zhang, T. Zhou, J. Wang, "Intelligent Fault Diagnosis Methods for Hydraulic Piston Pumps: A Review," Journal of Marine Science and Engineering, vol. 11 no. 3,DOI: 10.3390/JMSE11030594, 2023.

[6] H. S. Tang, Y. Ren, J. W. Xiang, "Fully-Coupled Thermomechanical Analysis for Sliding Contact Between Textured Slipper and Swashplate in Axial Piston Pump," International Journal of Heat and Mass Transfer, vol. 163,DOI: 10.1016/j.ijheatmasstransfer.2020.120521, 2020.

[7] Y. L. Chen, C. Z. Ban, Y. S. Liu, Y. Zhang, Y. Zhang, X. Q. Gao, "Wear Model and Life Prediction of Port Pair in Aircraft Piston Pump," Chinese Hydraulics & Pneumatics, vol. no. 12, 2020.

[8] Z. Du, Research on Fault Diagnosis of Axial Piston Pump Based on Variational Mode Decomposition and Multi Axis Fusion, 2022.

[9] L. Y. Li, "Analysis and Measure on Common Vibration and Noise in Hydraulic System," Hydraulics Pneumatics & Seals, vol. 31 no. 01, 2011.

[10] R. D. Lian, Research on Vibration Mechanism and Pattern Recognition Diagnosis Method of Axial Piston Pump, 2019.

[11] Z. J. Li, Study on Cavitation Detection and Fault Diagnosis of Axial Piston Pump, 2022.

[12] Y. Li, L. Y. Xu, F. Wang, S. Li, "Impact of Suction Pressure on Cavitation Characteristics in High-Speed Axial Piston Pump," Chinese Hydraulics & Pneumatics, vol. 49 no. 01, pp. 50-60, 2025.

[13] B. Zhang, Research on Vibration Mechanism and Vibration Evaluation of Hydraulic Pump, 2019.

[14] S. G. Ye, J. H. Zhang, B. Xu, L. Hou, J. W. Xiang, H. S. Tang, "A Theoretical Dynamic Model to Study the Vibration Response Characteristics of an Axial Piston Pump," Mechanical Systems and Signal Processing, vol. 150,DOI: 10.1016/j.ymssp.2020.107237, 2021.

[15] L. Q. Zhu, W. Y. Jiang, P. Li, B. Xing, "Research on the Fault Diagnosis Model for Movement Equipment Based on Identity Resolution and Credibility Matrix," Control and Instruments in Chemical Industry, vol. 47 no. 02, pp. 134-142, 2020.

[16] Q. Ma, D. Song, B. Liu, "Fault Mode Analysis and Simulation Verification of Hydraulic System Based on AMEsim," Journal of Physics: Conference Series, vol. 2006 no. 1,DOI: 10.1088/1742-6596/2006/1/012013, 2021.

[17] D. Bensaad, A. Soualhi, F. Guillet, "A New Leaky Piston Identification Method in an Axial Piston Pump Based on the Extended Kalman Filter," Measurement, vol. 148,DOI: 10.1016/j.measurement.2019.106921, 2019.

[18] Y. Shi, Research on Recognition of Wear Degradation State of Axial Piston Pump Based on Flow Field Analysis, 2021.

[19] T. C. Yang, J. Cai, Y. Huang, H. Y. Ma, "Modeling and Simulation of Swash Plate Axial Piston Pump Based on Modelica," Journal of Nanjing University of Aeronautics & Astronautics, vol. 54 no. 03, pp. 508-516, DOI: 10.16356/j.1005-2615.2022.03.018, 2022.

[20] S. Chen, Research on Fault Diagnosis of Piston Head Crack in Axial Piston Pump Based on Dynamic Simulation, 2022.

[21] J. C. Guo, Y. Liu, R. G. Yang, W. F. Sun, J. W. Xiang, "A simulation-Driven Difference Mode Decomposition Method for Fault Diagnosis in Axial Piston Pumps," Advanced Engineering Informatics, vol. 62,DOI: 10.1016/J.AEI.2024.102624, 2024.

[22] Z. J. Dai, Dynamic Analysis and Fault Characteristics of Piston Pair in Deep Sea Water Pump, 2023.

[23] H. B. Tang, J. Yang, Y. Tang, "Fatigue Damage Analysis and Life Prediction of Axial Piston Pump," Machine Tool & Hydraulics, vol. 51 no. 16, pp. 165-171, 2023.

[24] H. B. Tang, Z. X. Li, J. Y. Dong, S. Y. Chen, "Fault Diagnosis Method of Axial Piston Pump Based on Multi-Domain Coupling Modeling," Machine Tool & Hydraulics, vol. 52 no. 15, pp. 233-240, 2024.

[25] K. Przystupa, B. Ambrożkiewicz, G. Litak, "Diagnostics of Transient States in Hydraulic Pump System With Short Time Fourier Transform," Advances in Science and Technology Research Journal, vol. 14 no. 1, pp. 178-183, DOI: 10.12913/22998624/116971, 2020.

[26] C. Hu, Z. M. Li, "Research on Fault Diagnosis Method of Aviation Hydraulic Piston Pump Based on Multi-Characteristic Threshold Criteria," Modern Manufacturing Engineering, vol. no. 09, pp. 145-151, DOI: 10.16731/j.cnki.1671-3133.2023.09.019, 2023.

[27] G. F. Qi, D. Sun, W. B. Zheng, "Fault Diagnosis of Oil Field Water Injection Plunger Pump Based on Spectrum Analysis," Fluid Machinery, vol. 51 no. 03, pp. 84-90+98, 2023.

[28] S. Q. Xia, Y. M. Xia, J. W. Xiang, "Modelling and Fault Detection for Specific Cavitation Damage Based on the Discharge Pressure of Axial Piston Pumps," Mathematics, vol. 10 no. 14,DOI: 10.3390/MATH10142461, 2022.

[29] Z. Zheng, X. Z. Li, Y. Zhu, "Feature Extraction of the Hydraulic Pump Fault Based on Improved Autogram," Measurement, vol. 163,DOI: 10.1016/j.measurement.2020.107908, 2020.

[30] A. Hemati, A. Shooshtari, "Gear Pump Root Cause Failure Analysis Using Vibrations Analysis and Signal Processing," Journal of Failure Analysis and Prevention, vol. 20 no. 6, pp. 1815-1818, DOI: 10.1007/s11668-020-01008-3, 2020.

[31] J. Zhou, X. J. Ding, Y. F. Yang, T. Jing, Y. H. Wang, C. H. Wang, "Plunger Pump Cavitation Fault Recognition Based on Analysis of Low Frequency Energy," Global Reliability and Prognostics and Health Management,DOI: 10.1109/PHM-Nanjing52125.2021.9613023, 2021.

[32] A. Kumar, H. S. Tang, G. Vashishtha, J. W. Xiang, "Noise Subtraction and Marginal Enhanced Square Envelope Spectrum (MESES) for the Identification of Bearing Defects in Centrifugal and Axial Pump," Mechanical Systems and Signal Processing, vol. 165,DOI: 10.1016/J.YMSSP.2021.108366, 2022.

[33] H. Yu, H. R. Li, Y. L. Li, "Vibration Signal Fusion Using Improved Empirical Wavelet Transform and Variance Contribution Rate for Weak Fault Detection of Hydraulic Pumps," ISA Transactions, vol. 107, pp. 385-401, DOI: 10.1016/j.isatra.2020.07.025, 2020.

[34] Z. Zheng, Z. J. Wang, Y. Zhu, S. N. Tang, B. Z. Wang, "Feature Extraction Method for Hydraulic Pump Fault Signal Based on Improved Empirical Wavelet Transform," Processes, vol. 7 no. 11,DOI: 10.3390/pr7110824, 2019.

[35] S. C. Qiao, S. H. Xu, S. P. Wang, "Diagnosis Method of Slipper Wear Fault of Double-Oblique-Type Axia Piston Pump Based on Energy Enhancement," Chinese Hydraulics & Pneumatics, vol. 47 no. 02, pp. 62-71, 2023.

[36] Y. D. Wang, Y. Zhu, Q. L. Wang, S. Q. Yuan, S. N. Tang, Z. J. Zheng, "Effective Component Extraction for Hydraulic Pump Pressure Signal Based on Fast Empirical Mode Decomposition and Relative Entropy," AIP Advances, vol. 10 no. 7,DOI: 10.1063/5.0009771, 2020.

[37] S. Deng, L. W. Tang, X. J. Su, J. L. Che, "Fault Diagnosis Technology of Plunger Pump Based on EMMD-Teager," International Journal of Performability Engineering,DOI: 10.23940/ijpe.19.07.p18.19121919, 2019.

[38] C. A. Xiao, H. S. Tang, Y. Ren, J. W. Xiang, A. Kumar, "Adaptive MOMEDA Based on Improved Advance-Retreat Algorithm for Fault Features Extraction of Axial Piston Pump," ISA Transactions, vol. 128, pp. 503-520, DOI: 10.1016/J.ISATRA.2021.10.033, 2022.

[39] Z. Zheng, S. F. Li, Y. Guo, Z. J. Wang, Z. Chen, "Hydraulic Pump Fault Diagnosis Method UsingLog-SAM on Complex Signals," Journal of Vibration and Shock, vol. 40 no. 06, pp. 79-85, DOI: 10.13465/j.cnki.jvs.2021.06.010, 2021.

[40] H. R. Li, J. Sun, H. Ma, Z. K. Tian, Y. F. Li, "A Novel Method Based Upon Modified Composite Spectrum and Relative Entropy for Degradation Feature Extraction of Hydraulic Pump," Mechanical Systems and Signal Processing, vol. 114, pp. 399-412, DOI: 10.1016/j.ymssp.2018.04.040, 2019.

[41] S. Y. Liu, J. X. Yin, M. Hao, "Fault Diagnosis Study of Hydraulic Pump Based on Improved Symplectic Geometry Reconstruction Data Enhancement Method," Advanced Engineering Informatics, vol. 61,DOI: 10.1016/J.AEI.2024.102459, 2024.

[42] P. Casoli, M. Pastori, F. Scolari, M. Rundo, "A Vibration Signal-Based Method for Fault Identification and Classification in Hydraulic Axial Piston Pumps," Energies, vol. 12 no. 5,DOI: 10.3390/en12050953, 2019.

[43] H. B. Tang, Z. Fu, Y. Huang, "A Fault Diagnosis Method for Loose Slipper Failure of Piston Pump in Construction Machinery Under Changing Load," Applied Acoustics, vol. 172,DOI: 10.1016/j.apacoust.2020.107634, 2021.

[44] C. G. Buiges, C. König, "A Sensor Data-Based Approach for the Definition of Condition Taxonomies for a Hydraulic Pump," Engineering Proceedings, vol. 2 no. 1,DOI: 10.3390/ECSA-7-08223, 2020.

[45] L. C. Gu, Z. W. Ma, Q. Q. Tian, Y. Sun, "Application of Instantaneous Speed Fluctuation Signal in Fault Diagnosis of Axial Piston Pump," Journal of Drainage and Irrigation Machinery Engineering, vol. 39 no. 07, pp. 740-746, 2021.

[46] L. H. Zhao, W. Y. Li, Y. Cheng, C. Guan, "Research on Fault Diagnosis Method of Plunger Pump Based on LMD and Support Vector Machine," Machinery Design & Manufacture, vol. no. 03, pp. 238-241, DOI: 10.19356/j.cnki.1001-3997.20211115.016, 2022.

[47] Q. Chao, Y. C. Shao, C. L. Liu, X. X. Yang, "Health Evaluation of Axial Piston Pumps Based on Density Weighted Support Vector Data Description," Reliability Engineering & System Safety, vol. 237,DOI: 10.1016/J.RESS.2023.109354, 2023.

[48] W. Y. Li, Y. Cheng, L. H. Zhao, L. Han, "Shoe Wear Fault Diagnosis of Axial Piston Pump Based on Local Transform and Extreme Learning Machine," Chinese Hydraulics & Pneumatics, vol. no. 12, pp. 15-21, 2019.

[49] Y. Cheng, W. Y. Li, L. Quan, L. H. Zhao, C. Guan, L. Han, "Weak Fault Diagnosis of Axial Piston Pump Based on VMD-MDE and ELM," Journal of Vibration, Measurement & Diagnosis, vol. 40 no. 04, pp. 635-642+818, DOI: 10.16450/j.cnki.issn.1004-6801.2020.04.001, 2020.

[50] W. L. Jiang, Z. B. Li, J. J. Li, Y. Zhu, P. Y. Zhang, "Study on a Fault Identification Method of the Hydraulic Pump Based on a Combination of Voiceprint Characteristics and Extreme Learning Machine," Processes, vol. 7 no. 12,DOI: 10.3390/pr7120894, 2019.

[51] Z. B. Li, W. L. Jiang, S. Zhang, Y. Sun, S. Q. Zhang, "A Hydraulic Pump Fault Diagnosis Method Based on the Modified Ensemble Empirical Mode Decomposition and Wavelet Kernel Extreme Learning Machine Methods," Sensors, vol. 21 no. 8,DOI: 10.3390/S21082599, 2021.

[52] Y. Ding, L. Ma, C. Wang, L. F. Tao, "An EWT-PCA and Extreme Learning Machine Based Diagnosis Approach for Hydraulic Pump," IFAC-PapersOnLine, vol. 53 no. 3, pp. 43-47, DOI: 10.1016/J.IFACOL.2020.11.008, 2020.

[53] G. H. Lee, U. E. Akpudo, J. W. Hur, "FMECA and MFCC-Based Early Wear Detection in Gear Pumps in Cost-Aware Monitoring Systems," Electronics, vol. 10 no. 23,DOI: 10.3390/ELECTRONICS10232939, 2021.

[54] Y. P. Zhang, W. H. Tian, Y. H. Liu, "Fault Diagnosis of Slipper Wear Signal of Piston Pump Based on RandomForest Algorithm," Machinery Design & Manufacture, vol. no. 06, pp. 150-153, DOI: 10.19356/j.cnki.1001-3997.20230724.031, 2024.

[55] X. Y. Ma, The Method of Imbalanced Data Processing and Its Application Research in Fault Diagnosis of Axial Piston Pump, 2022.

[56] W. L. Jiang, X. Y. Ma, Y. Yue, Y. P. Zhao, "Fault Diagnosis Method of Axial Piston Pump Based on Balanced Random Forest Under Lmbalanced Datasets," Chinese Hydraulics & Pneumatics, vol. 46 no. 03, pp. 45-54, 2022.

[57] Y. Z. Jia, M. Q. Xu, R. X. Wang, "Symbolic Important Point Perceptually and Hidden Markov Model Based Hydraulic Pump Fault Diagnosis Method," Sensors, vol. 18 no. 12,DOI: 10.3390/s18124460, 2018.

[58] W. L. Jiang, Z. B. Li, S. Zhang, Y. F. Lei, H. N. Wang, "Fault Recognition Method Based on Recurrence Quantitation Analysis for Hydraulic Pump," Chinese Hydraulics & Pneumatics, vol. 2019 no. 02, pp. 18-23, .

[59] W. L. Jiang, P. Y. Zhang, M. Li, S. Q. Zhang, "Axial Piston Pump Fault Diagnosis Method Based on Symmetrical Polar Coordinate Image and Fuzzy-C-Means Clustering Algorithm," Shock and Vibration, vol. 2021 no. 1,DOI: 10.1155/2021/6681751, 2021.

[60] L. C. Jin, J. K. Ye, Z. Zhang, X. M. Tang, X. Y. Shao, S. Tuo, "Fault Diagnosis of Hydraulic Pump Based on DMD and t-SNE," Machine Tool & Hydraulics, vol. 49 no. 14, pp. 187-192+200, 2021.

[61] L. Y. Wang, Y. H. Wang, L. H. Meng, "Identification of Cavitation Intensity of High-Speed Axial Piston Pumps Based on Variational Mode Decomposition and XGBoost," Chinese Hydraulics & Pneumatics, vol. 45 no. 05, pp. 62-67, 2021.

[62] X. Fu, "Bayesian Network Based Fault Diagnosis of Aero Hydraulic Pump," CSAA/IET International Conference on Aircraft Utility Systems (AUS 2020), pp. 539-543, DOI: 10.1049/icp.2021.0454, .

[63] Z. D. Du, J. M. Zhao, H. P. Li, X. Zhang, "A Fault Diagnosis Method of a Plunger Pump Based on SA-EMD-PNN," Journal of Vibration and Shock, vol. 38 no. 08, pp. 145-152, DOI: 10.13465/j.cnki.jvs.2019.08.022, 2019.

[64] X. L. Wei, Q. Chao, J. F. Tao, C. L. Liu, L. Y. Wang, "Cavitation Fault Diagnosis of High-Speed Axial Piston Pump Based on Spectrum Analysis and Convolutional Neural Network," Chinese Hydraulics & Pneumatics, vol. 45 no. 07, 2021.

[65] Z. J. Wu, C. A. Xiao, H. S. Tang, Y. Ren, Y. He, "Fault Diagnosis of Axial Piston Pump Based on Polynomial chirplet Transform and Variational Mode Decomposition Under Variable Speed Conditions," Chinese Hydraulics & Pneumatics, vol. 45 no. 07, pp. 77-82, 2021.

[66] J. Q. Teng, F. Luo, Q. Zhang, Y. T. Zhang, T. B. Xia, "A Sensors Location Considered Piston Pump Fault Diagnosis Model With Bagging Based Convolutional Neural Network," Machine Design and Research, vol. 40 no. 04, pp. 220-225, DOI: 10.13952/j.cnki.jofmdr.2024.0167, 2024.

[67] L. W. Chen, P. T. Ying, H. S. Tang, Y. Ren, J. W. Xiang, "Fault Diagnosis of Axial Piston Pump Based on Multi-Sensor Data Fusionand Deep Residual Shrinkage Network," Chinese Hydraulics & Pneumatics, vol. 47 no. 11, pp. 142-149, 2023.

[68] J. Q. Fan, Y. Lan, J. H. Huang, X. Y. Xiong, G. Y. Li, L. N. Li, "Fault Diagnosis Modal of Axial Piston Pump Based on Prototypical Network With Small Sample Size," Journal of Mechanical & Electrical Engineering, vol. 40 no. 04, pp. 584-591, 2023.

[69] M. N. Hu, Research on Fault Diagnosis and Life Prediction Technology of Plunger Pump Based on Graph Neural Network, 2023.

[70] L. H. Meng, M. H. Zhao, Z. Q. Cui, X. M. Zhang, S. S. Zhong, "Empirical Mode Reconstruction: Preserving Intrinsic Components in Data Augmentation for Intelligent Fault Diagnosis of Civil Aviation Hydraulic Pumps," Computers in Industry, vol. 134,DOI: 10.1016/J.COMPIND.2021.103557, 2022.

[71] H. H. Gao, Q. Chao, Z. Xu, J. F. Tao, M. Y. Liu, C. L. Liu, "Piston Pump Fault Diagnosis Based on Siamese Neural Network With Small Samples," Journal of Beijing University of Aeronautics and Astronautics, vol. 49 no. 01, pp. 155-164, DOI: 10.13700/j.bh.1001-5965.2021.0213, 2023.

[72] Y. Zhang, Y. He, H. Tang, Y. Ren, J. Xiang, "Adversarial Domain Adaptation Approach for Axial Piston Pump Fault Diagnosis Under Small Sample Condition Based on Measured and Simulated Signals," IEEE Transactions on Instrumentation and Measurement, vol. 73,DOI: 10.1109/tim.2024.3385829, 2024.

[73] H. D. Ma, Research on Hydraulic Pump Fault Diagnosis Method Based on Big Datacloud Platform and Deep Learning Network CNN, 2021.

[74] D. Wang, S. Liu, W. Huang, J. Zhang, B. Xu, "An Online Wear State Identification Method for Axial Piston Pump Key Friction Pair Based on FSANN," 2022 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD),DOI: 10.1109/icsmd57530.2022.10058342, 2022.

[75] D. D. Wang, K. F. Miao, F. Lv, J. H. Zhang, S. G. Liu, W. D. Huang, "A Lightweight Method for Online Wear State Recognition of Key Friction Pairs in Axial Piston Pump," IEEE Transactions on Industrial Informatics, vol. 48 no. 10, 2024.

[76] Y. Zhu, G. P. Li, S. N. Tang, R. Wang, H. Su, C. Wang, "Acoustic Signal-Based Fault Detection of Hydraulic Piston Pump Using a Particle Swarm Optimization Enhancement CNN," Applied Acoustics, vol. 192,DOI: 10.1016/J.APACOUST.2022.108718, 2022.

[77] C. L. Wang, J. J. Sun, Y. T. Zhong, "Fault Intelligent Identification Method for Axial Piston Pump Using Multi-Subarray Acoustical Signal," Manufacturing Technology & Machine Tool, vol. no. 09, pp. 23-28, DOI: 10.19287/j.mtmt.1005-2402.2024.09.003, 2024.

[78] P. Y. Zhang, Research on Piston Pump Fault Diagnosis Method Basedon Voiceprint Image and Deep Transfer Learning, 2024.

[79] Y. Liu, Y. S. Dai, L. G. Li, "Fault Diagnosis of Plunger Pumps Based on Audio Signal of Small Samples," Noise and Vibration Control, vol. 43 no. 05, pp. 142-147+273, 2023.

[80] W. Li, Y. Liu, L. G. Li, L. Zhou, C. S. Song, "Audio Feature Extraction Method for Plunger Pump Fault Based on Adaptive Noise Reduction," Foreign Electronic Measurement Technology, vol. 42 no. 01,DOI: 10.19652/j.cnki.femt.2204291, 2023.

[81] Z. Xu, Q. Chao, H. H. Gao, J. F. Tao, C. L. Liu, C. W. Meng, "A Fault Diagnosis Method for Piston Pump Under Variable Speed Conditions Using Parameterized Demodulation," Journal of Xi’an Jiaotong University, vol. 55 no. 10, pp. 19-29, 2021.

[82] Z. Y. Wang, Z. Zhou, W. G. Xu, C. Sun, R. Q. Yan, "Physics Informed Neural Networks for Fault Severity Identification of Axial Piston Pumps," Journal of Manufacturing Systems, vol. 71, pp. 421-437, DOI: 10.1016/J.JMSY.2023.10.002, 2023.

[83] S. N. Tang, Y. Zhu, S. Q. Yuan, "A Novel Adaptive Convolutional Neural Network for Fault Diagnosis of Hydraulic Piston Pump With acoustic Images," Advanced Engineering Informatics, vol. 52,DOI: 10.1016/J.AEI.2022.101554, 2022.

[84] S. N. Tang, Y. Zhu, S. Q. Yuan, "Intelligent Fault Diagnosis of Hydraulic Piston Pump Based on Deep Learning and Bayesian Optimization," ISA Transactions, vol. 129, pp. 555-563, DOI: 10.1016/J.ISATRA.2022.01.013, 2022.

[85] S. N. Tang, Y. Zhu, S. Q. Yuan, "Intelligent Fault Identification of Hydraulic Pump Using Deep Adaptive Normalized CNN and Synchrosqueezed Wavelet Transform," Reliability Engineering & System Safety, vol. 224,DOI: 10.1016/J.RESS.2022.108560, 2022.

[86] Y. Zhu, G. P. Li, R. Wang, S. N. Tang, H. Su, K. Cao, "Intelligent Fault Diagnosis of Hydraulic Piston Pump Based on Wavelet Analysis and Improved AlexNet," Sensors, vol. 21 no. 2,DOI: 10.3390/S21020549, 2021.

[87] Y. Zhu, S. N. Tang, S. Q. Yuan, "Multiple-Signal Defect Identification of Hydraulic Pump Using an Adaptive Normalized Model and S Transform," Engineering Applications of Artificial Intelligence, vol. 124,DOI: 10.1016/J.ENGAPPAI.2023.106548, 2023.

[88] Z. Y. Wang, T. F. Li, W. G. Xu, "Denoising Mixed Attention Variational Auto-Encoder for Axial Piston Pump Fault Diagnosis," Journal of Mechanical Engineering, vol. 60 no. 04, pp. 167-177, 2024.

[89] W. L. Jiang, Z. B. Li, S. Zhang, T. Wang, S. Q. Zhang, "Hydraulic Pump Fault Diagnosis Method Based on EWT Decomposition Denoising and Deep Learning on Cloud Platform," Shock and Vibration, vol. 2021 no. 1,DOI: 10.1155/2021/6674351, 2021.

[90] S. F. Cui, H. Q. Song, F. Li, Z. G. Lu, "Multi-Source Fault Signal Fusion Diagnosis of Hydraulic Pump Based on PSO-BP and D-S Evidence," Machine Design and Research, vol. 38 no. 02, pp. 155-157+173, DOI: 10.13952/j.cnki.jofmdr.2022.0025, 2022.

[91] G. L. Zhu, Z. Q. Wang, Y. F. Wang, Z. F. Li, C. Z. Sun, "Hydraulic Pump Fault Diagnosis Based on Neural Network and Evidence Fusion," Journal of Mechanical & Electrical Engineering, vol. 37 no. 12, pp. 1498-1503, 2020.

[92] Q. Chao, H. H. Gao, J. F. Tao, Y. H. Wang, J. Zhou, C. L. Liu, "Adaptive Decision-Level Fusion Strategy for the Fault Diagnosis of Axial Piston Pumps Using Multiple Channels of Vibration Signals," Science China Technological Sciences, vol. 65 no. 2, pp. 470-480, DOI: 10.1007/s11431-021-1904-7, 2022.

[93] B. Li, "Fault Diagnosis Method of Ship Hydraulic System Based on KPCA-PNN," 2021 International Conference on Networking, Communications and Information Technology (NetCIT), pp. 73-77, 2021.

[94] P. F. Liang, X. F. Wang, C. Ai, D. M. Hou, S. Y. Liu, "SRSGCN: A Novel Multi-Sensor Fault Diagnosis Method for Hydraulic Axial Piston Pump With Limited Data," Reliability Engineering & System Safety, vol. 253,DOI: 10.1016/J.RESS.2024.110563, 2025.

[95] S. Liu, J. Zhang, W. Huang, F. Lyu, D. Wang, B. Xu, "Temporal–Spatial Attention Network: A Novel Axial Piston Pump Coupled Fault Diagnosis Method," IEEE Transactions on Instrumentation and Measurement, vol. 73,DOI: 10.1109/tim.2024.3398074, 2024.

[96] Z. K. Dong, H. J. An, S. Y. Liu, "Domain Adversarial Transfer Fault Diagnosis Method of an Axial Piston Pump Based on a Multi-Scale Attention Mechanism," Measurement, vol. 239,DOI: 10.1016/J.MEASUREMENT.2024.115455, 2025.

[97] X. L. Xu, J. H. Zhang, W. D. Huang, "The Loose Slipper Fault Diagnosis of Variable-Displacement Pumps Under Time-Varying Operating Conditions," Reliability Engineering & System Safety, vol. 252,DOI: 10.1016/J.RESS.2024.110448, 2024.

[98] Y. He, H. N. Tang, Y. Ren, A. Kumar, "A Deep Multi-Signal Fusion Adversarial Model Based Transfer Learning and Residual Network for Axial Piston Pump Fault Diagnosis," Measurement, vol. 192,DOI: 10.1016/J.MEASUREMENT.2022.110889, 2022.

[99] Y. Miao, Y. C. Jiang, J. F. Huang, X. J. Zhang, L. Han, "Application of Fault Diagnosis of Seawater Hydraulic Pump Based on Transfer Learning," Shock and Vibration, vol. 2020 no. 1,DOI: 10.1155/2020/9630986, 2020.

[100] H. Y. Qiu, C. F. Zhang, B. Xu, Q. Su, R. L. Wang, "Fault Diagnosis of Hydraulic Drive System of Rapid-Erection Device Based on Optimized BP Neural Network," Chinese Hydraulics & Pneumatics, vol. 45 no. 03, 2021.

[101] J. L. Zhang, L. X. Duan, Z. X. Wang, W. Q. Wang, "Research on Fault Diagnosis Method of Fracturing Pump Based on K-TrAdaBoost Transfer Learning," China Measurement & Test, vol. 47 no. 10, 2021.

[102] Z. J. Li, Y. Lan, J. H. Huang, "Cavitation State Detection of Axial Piston Pump Based on CBLRE Model," Journal of Mechanical & Electrical Engineering, vol. 39 no. 05, pp. 634-664, 2022.