An Explainable Artificial Intelligence-Based Framework for Diagnosing Faults in an Induction Furnace

For over a century, induction furnaces have been used in the core of foundries for metal melting and heating. They provide high melting/heating rates with an optimal efficiency. Occurrence of faults not only impose safety risks but also reduce productivity due to unscheduled shutdowns. The problem of diagnosing faults in induction furnaces (IFs) has not been studied yet and this is the first such work, which proposes a data-driven framework for diagnosing faults in IFs. We propose a deep neural network that detects electrical faults by measuring real-time electrical parameters at the supply side. Experimental and sensory measurements are collected from multiple energy analyzer devices installed in a foundry. Next, a semi-supervised learning algorithm, called Local Outlier Factor, has been used to discriminate normal and faulty samples from each other. Then, a deep neural network is trained with collected samples. The performance of the developed model is compared with several state-of-the-art techniques in terms of various performance metrics. Due to the black-box nature of the DNN, the model prediction are interpreted by SHAP and LIME methods. Instead of only providing reasons for a specific decision, we have used LIME on individual faults iteratively to obtain global explanations together with SHAP interpretations. Using the global explanations, we identify electrical parameters contributing most to each fault class.