Industrial

In the bearings’ segment of machine tools, there is a strong demand for high-performance steel solutions. The bearings may operate under severe conditions of contact pressures of up to 3 GPa; rotating at speed factors in excess of 3 million ndm. Such conditions pose a high risk of bearing seizure failure. Improving lubrication conditions is complex as the bearing operating temperature must be well controlled. Adhesive wear was found to occur in hybrid steel-ceramic contacts. Another relevant failure mode is micropitting. It is demonstrated that macroscopic hardness is insufficient to predict the resistance of steel microstructures to surface-initiated fatigue. In this regard, strain-hardening and the breadth of the range of hardness values of microstructure phases play an important role.

Wear is a type of surface damage commonly observed in industrial components in relative motion and in contact with other solid surfaces. The majority of wear occurs progressively in a given contact starting from an initial running-in period followed by a steady-state period. Being able to accurately classify the running-in and steady-state periods allow reducing significant production or damage costs of complex machines, in particular when the load varies during operation. Production cost can be addressed by optimizing the running-in time. In contrast, significant damages can takes place if the machine are of set to full production capacity before the running-in time is finished. To address these two problems, we use a real-time monitoring system to differentiate between running-in and steady-state periods as well as classify the loading conditions simultaneously based on AE signals using a multi-label Convolutional Neural Network (CNN). Reciprocating sliding tests are performed at two loads (200 and 500 g). The tribopair used is a steel ball sliding against steel plates under dry conditions. The tribotest is divided into two different states, running-in and steady-state based on the obtained friction curves. A pico-acoustic sensor is attached on the steel plate's surface, the fix body, to acquire AE signals during the friction test. Raw AE signals are processed and directly analyzed using a multi-label CNN to simultaneously classify the running-in and steady-state periods as well as the loading conditions. This machine learning method accurately classifies the running-in and steady-state as well as the loading conditions with a 99% average accuracy.