A lightweight model compression framework for intelligent fault diagnosis of machines on resource-constrained devices

With the increasing complexity and automation of industrial equipment, deploying mechanical fault diagnosis systems on edge devices is limited by computing and storage resources, necessitating model optimisation to reduce these requirements. To address these limitations, this paper proposes a model compression framework for mechanical fault diagnosis aimed at reducing computational complexity and improving deployment efficiency on edge devices. Initially, the vibration signals are preprocessed and converted into Gram Angle Field (GAF) images to retain essential time-domain features. Subsequently, a filter pruning strategy is applied to reduce model parameters, eliminate redundancy, and simplify the model structure. Furthermore, the model is quantised to further lower the storage requirements and computational complexity during inference. Experimental results show that the proposed model compression framework reduces the size of pre-trained mechanical fault diagnosis model by 94.35% without compromising accuracy. This significant reduction in computational cost and storage requirements makes the approach a feasible solution for implementing mechanical fault diagnosis on edge devices.