Prognostic health management of the robotic strain wave gear reducer based on variable speed of operation: a data-driven via deep learning approach

Abstract

The robotic reducer is prone to failure because of its unique characteristics. Data from vibration and acoustic emission sensors have been used for the prognostics of the reducer. However, various issues are associated with such traditional techniques. Hence, our research group proposes a novel approach to utilize the embedded setup of the electrical current to detect the mechanical fault of the robotic reducer in the actual industrial robot. Previously, a comprehensive approach of feature engineering was proposed to classify the mechanical fault for the robotic reducer. However, handcraft-based feature extraction is quite a tedious task, and computationally expensive. These features require a well-designed feature extractor, and the features need to be manually optimized before feeding into classifiers. In addition, the handcrafted features are problem-specific, and are complicated to generalize. To resolve these challenges, deep features are extracted to classify the fault and generalize for two different motion profiles under different working conditions. In the proposed research work, the fault characteristic is generalized for variable speed of operations considering various kinds of scenarios. In this research work, the generalization capability of the proposed approach is comprehensively evaluated. For that purpose, the data under different working conditions such as of lower speeds, higher speeds, and speed sequestration are used as unseen data to validate the model. The authenticity of the presented approach can be supported by the performance evaluation for fault classification of the different motion profiles and speed of operations.