Relating Process Data to Product Quality in Transmission Manufacturing

Abstract: Research has shown that vibration produced by a machine tool provides useful information for the maintenance of its structural parts. Thus, the vibration signal should also give information about tool wear and, consequently, surface roughness of the machined workpieces. Tool wear may cause defects in the machined components, thus a conservative approach is often taken, replacing inserts before they should have been. Monitoring the status of the tool in real-time would provide a valuable way to perform condition-based maintenance and, once a trend is discovered, to perform predictive maintenance. This has the advantages that the asset lifespan is increased, tool change is performed only when needed and the health of the tool has a real time-indication. Literature study and a benchmarking analysis have been performed, in order to understand the state-of-the-art regarding vibration analysis and product quality prediction. Tests have been carried out in order to verify the feasibility of vibration signal use to monitor toolwear and surface roughness in hard turning. The machined workpieces are crown wheels used for the transmission of trucks in Scania and the tools are in polycrystalline cubic boron nitride (PCBN). The vibration signals are recorded by two tri-axial accelerometers, the tool inserts are examined by microscope and the machined parts are measured both with a CMM machine and a profilometer. The analysis of the vibration data includes pre-processing to filter the signals, understanding the vibration metrics such as root mean square, kurtosis, crest factor and peak, regression analysis to try to predict the surface roughness. The results show that the vibration metrics, in particular kurtosis, crest factor and root mean square have a clear trend showing the increase in tool wear, which can be a useful indicator of surface roughness. Moreover, it is found out that regression analysis using vibration signal features as predictors and any of the relevant dimensional accuracy value as the response does not provide satisfactory results. On the other hand, regression analysis using vibration signal metrics as predictors and one of the surface roughness parameters (Ra or Rq) as the response can provide a good model to predict surface quality. Since the model is obtained with a limited dataset, it will be necessary to validate it with further tests. Future work would consist in providing a long-term solution installing permanent sensors, finding a way to automatically predict surface roughness on-line and adapting this project to fit other machines and parts.