Can machine fault diagnosis be conducted using the voltage outputs of a nonlinear piezoelectric energy harvester?

Abstract Nonlinear piezoelectric energy harvesting (PEH) system shows great potential in self-powered sensors for machine status monitoring. Recently, it is supposed to use the voltage outputs of a PEH for machine fault diagnosis as named ‘self-sensing’. However, whether the nonlinear dynamic responses convey sufficient diagnostic information in vibration signals has not been fully explored as nonlinearity would modulate the responses. To meet the dual challenges of simultaneous energy harvesting and sensing, this study comprehensively evaluates the feasibility and limitations of using nonlinear PEH for bearing fault diagnosis. A dynamic model for monostable and bistable PEH is established. Real Case Western Reserve University datasets for fault bearing is used as the excitation signals, and the electrical outputs of the nonlinear PEH were evaluated. In addition, a fault diagnosis framework based on logarithm spectrum characteristics was developed, combining a one-dimensional convolutional neural network and a traditional machine learning classifiers to identify the location and severity of bearing faults. The results demonstrate that nonlinear PEH will produce modulation in time domain as well as frequency domain. Despite the higher energy harvesting efficiency of a bistable condition, it introduces more complex nonlinear distortion than the monostable one. Nevertheless, the proposed framework can achieve high fault classification accuracy under both working conditions. These findings provide theoretical guidance for optimizing the nonlinear structure of the self-powered fault monitoring system and balance the relationship between energy generation and diagnostic fidelity.