Knowledge Transfer based Evolutionary Deep Neural Network for Intelligent Fault Diagnosis

TL;DR

The paper tackles fault diagnosis under changing operating conditions by addressing two key challenges: selecting an appropriate neural network architecture and achieving domain adaptation with limited labeled target data. It introduces EvoN2N, an evolutionary Net2Net framework that uses NSGA-II for multi-objective architecture search, and employs Net2Net-based quick learning to transfer knowledge between generations, coupled with MMD-based domain alignment. The method jointly optimizes accuracy and model compactness while rapidly adapting to new data domains, validated on CWRU, PBU, and Gearbox Fault datasets, achieving high diagnostic performance and statistically significant improvements over baselines. The approach provides a practical path to real-time, robust fault diagnosis in industrial settings with limited data, thanks to transfer-enabled architecture evolution and efficient fitness evaluation.

Abstract

A faster response with commendable accuracy in intelligent systems is essential for the reliability and smooth operations of industrial machines. Two main challenges affect the design of such intelligent systems: (i) the selection of a suitable model and (ii) domain adaptation if there is a continuous change in operating conditions. Therefore, we propose an evolutionary Net2Net transformation (EvoN2N) that finds the best suitable DNN architecture with limited availability of labeled data samples. Net2Net transformation-based quick learning algorithm has been used in the evolutionary framework of Non-dominated sorting genetic algorithm II to obtain the best DNN architecture. Net2Net transformation-based quick learning algorithm uses the concept of knowledge transfer from one generation to the next for faster fitness evaluation. The proposed framework can obtain the best model for intelligent fault diagnosis without a long and time-consuming search process. The proposed framework has been validated on the Case Western Reserve University dataset, the Paderborn University dataset, and the gearbox fault detection dataset under different operating conditions. The best models obtained are capable of demonstrating an excellent diagnostic performance and classification accuracy of almost up to 100% for most of the operating conditions.

Figures (7)

• Figure 1: Domain shift between source and target dataset: A slight change in the data distribution from the source domain (blue) to the target domain (red) causes miss-classification and needs fine-tuning of the model for domain adjustment.
• Figure 2: Building blocks of NSGA-II frameworks for neural network architecture optimization using N2N transformation-based quick learning mechanism.
• Figure 3: Fitness evaluation strategy.
• Figure 4: Crossover of two different length chromosomes: (a) $n_i$ = number of hidden layers in $i^{th}$ chromosome (model), $h_{ij}$ = number of hidden nodes in $j^{th}$ hidden layer of $i^{th}$ model, $\bar{n}_i$ = number of hidden layers in $i^{th}$ offspring model, and $\bar{h}_{ij}$ = number of hidden nodes in $j^{th}$ hidden layer of $i^{th}$ offspring model; (b) Illustration of crossover with numerical values (values are taken for illustration only).
• Figure 5: Growth of CA with generation for the first load condition from each case.