Abstract:Deep learning (DL) has gained popularity in recent years as an effective tool for classifying the current health and predicting the future of industrial equipment. However, most DL models have black-box components with an underlying structure that is too complex to be interpreted and explained to human users. This presents significant challenges when deploying these models for safety-critical maintenance tasks, where non-technical personnel often need to have complete trust in the recommendations these models give. To address these challenges, we utilize a convolutional neural network (CNN) with Gradient-weighted Class Activation Mapping (Grad-CAM) activation map visualizations to form an interpretable DL method for classifying bearing faults. After the model training process, we apply Grad-CAM to identify a training sample's feature importance and to form a library of diagnosis knowledge (or health library) containing training samples with annotated feature maps. During the model evaluation process, the proposed approach retrieves prediction basis samples from the health library according to the similarity of the feature importance. The proposed method can be easily applied to any CNN model without modifying the model architecture, and our experimental results show that this method can select prediction basis samples that are intuitively and physically meaningful, improving the model's trustworthiness for human users.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the interpretability of deep - learning models in industrial equipment health monitoring and fault diagnosis. Specifically, most deep - learning (DL) models are regarded as "black boxes" due to their complex internal structures, and it is difficult to explain their decision - making processes to users. This is particularly prominent in scenarios involving safety - critical tasks, such as the fault diagnosis of rotating equipment (such as bearings), because these tasks require non - technical users to have complete trust in the model's recommended results. ### Core Contributions of the Paper 1. **Propose an interpretable deep - learning method**: - The authors use a convolutional neural network (CNN) combined with Gradient - weighted Class Activation Mapping (Grad - CAM) to generate interpretable model predictions, enabling human users to understand the model's predictions of bearing health. 2. **Two - step procedure**: - **Health library creation**: Generate activation vectors by calculating the Grad - CAM activation maps of each training sample, and store them together with the predicted health status in the health library. - **Prediction - base - sample retrieval**: During the model evaluation process, select the most similar training sample as the prediction - base - sample according to the feature - importance similarity between the test sample and all training samples in the health library. 3. **Improve model credibility**: - This method can be applied to any CNN model without modifying its architecture, and the experimental results show that the selected prediction - base - samples have intuitive and physical significance, thereby increasing the model's credibility to users. ### Formula Summary - **Bearing fault characteristic frequency formula**: \[ f_{BPFO} = \frac{n \times f_r}{2} \left[1 - \frac{d}{D} \times \cos(\phi)\right] \] \[ f_{BPFI} = \frac{n \times f_r}{2} \left[1 + \frac{d}{D} \times \cos(\phi)\right] \] - **Grad - CAM activation map weight calculation**: \[ \alpha^c_k = \frac{1}{C} \sum_i \frac{\partial y_c}{\partial A^k_i} \] \[ L^c_{\text{Grad - CAM}} = \sum_k \alpha^c_k A^k \] - **Normalized class - activation - vector calculation**: \[ \text{TestCANorm}^c_j = \frac{\text{TestCA}^c_j}{||\text{TestCA}^c_j||} \] \[ \text{TrainCANorm}^c_i = \frac{\text{TrainCA}^c_i}{||\text{TrainCA}^c_i||} \] - **Euclidean - distance calculation**: \[ \text{Dis}_{i,j} = ||\text{TestCANorm}^c_i - \text{TrainCANorm}^c_j|| \] Through these methods, the paper provides an effective and interpretable deep - learning framework for bearing - fault diagnosis, enabling non - technical personnel to understand and trust the model's prediction results.

An interpretable deep learning method for bearing fault diagnosis

Fault Diagnosis Method Based on Improved Deep Boltzmann Machines

Long-term evaluation of modified Tanzer ear reconstruction.

Interpretability of deep convolutional neural networks on rolling bearing fault diagnosis

A Deep Learning Method for Bearing Fault Diagnosis Based on Time-Frequency Image

A Lightweight Model for Bearing Fault Diagnosis Based on Gramian Angular Field and Coordinate Attention

Hierarchical adaptive deep convolution neural network and its application to bearing fault diagnosis

Bearing fault diagnosis method based on CNN-LightGBM

Intelligent Fault Diagnosis of Bearing Based on Convolutional Neural Network and Bidirectional Long Short-Term Memory

An Improved Hybrid CNN-SVM Based Method for Bearing Fault Diagnosis under Noisy Environment

Deep Learning Algorithms for Bearing Fault Diagnostics—A Comprehensive Review

A multi-scale collaborative fusion residual neural network-based approach for bearing fault diagnosis

A novel two-stream multi-head self-attention convolutional neural network for bearing fault diagnosis

Deep transfer learning rolling bearing fault diagnosis method based on convolutional neural network feature fusion

Intelligent fault diagnosis of roller bearing based on dual-flow convolutional neural network

A New Convolutional Neural Network-Based Data-Driven Fault Diagnosis Method

Rolling bearing fault diagnosis model based on DSCB-NFAM

A New Interpretable Learning Method for Fault Diagnosis of Rolling Bearings

Bearing fault diagnosis with parallel CNN and LSTM

Review of the U.S. military's human immunodeficiency virus program: a legacy of progress and a future of promise.