Bolt Torque Assessment in Automotive Connecting Rod using Contact Acoustic Nonlinearity and Machine Learning

Ensuring that the bolts of an automotive connecting rod retain the correct preload is essential for the reliability of the entire assembly. However, widely used tightening or inspection techniques, ranging from torque-based procedures to conventional ultrasonic checks on individual bolts, often struggle with accuracy because of friction variability, and they offer no real means of monitoring the joint while the component is in service. This makes it difficult to detect early stages of loosening or changes in the contact conditions at the interface. In this proof-of-concept work, carried out using a real automotive component and in laboratory conditions, we explore a different route. Using contact acoustic nonlinearity, we examine how the nonlinear ultrasonic response of the connecting rod changes under different tightening torques. Measurements were carried out with a pitch–catch arrangement and concentrated around a frequency (18.3 kHz) identified experimentally as a local damage resonance, where the sensitivity to contact-related phenomena is noticeably enhanced. From these signals, several nonlinear indicators, most notably the amplitudes of the second and third harmonics, were extracted and then used to train a set of supervised machine-learning regressors aimed at estimating the applied torque in an automatic way. The results point to a consistent, monotonic link between the nonlinear parameters and the bolt preload. Among the tested regressors, Gradient Boosting achieved the most reliable predictions. Taken together, these findings suggest that combining nonlinear ultrasonics with data-driven models may offer a practical pathway towards non-invasive, possibly real-time monitoring of bolted joints in mechanically demanding environments.