Abstract: How to obtain a large number of fault samples from mechanical transmission systems under the actual running state is a bottleneck for the engineering application using intelligent diagnosis methods. To meet the requirement of precision diagnosis for individual differences, the basic principle of personalized diagnosis for mechanical faults is proposed. Through the construction of numerical simulation model of mechanical transmission systems, simulations are performed to obtain fault samples. The bottleneck problem of lacking fault feature information in the diagnostic procedures will be resolved to activate the artificial intelligent (AI) diagnosis methods. Taking the bearing, gear transmission, rotor-bearing system for examples, the finite element method (FEM) models of intact structures are firstly constructed to obtain simulation model with a certain precision using model updating techniques. Secondly, predefined several faults and further inserted into the high fidelity FEM model to calculate the fault samples, which severed as training samples of AI diagnostic models to classify testing samples (faults to be diagnosed). Finally, the experimental investigations using the arbitrary selection of support vector machine (SVM), extreme learning machine (ELM) and convolutional neural network (CNN) show that the principle of personalized diagnosis for mechanical faults has strong universality and expansibility.

Key words: mechanical transmission systems, faults, personalized diagnosis, artificial intelligent models, classification