Mold Damage Monitoring for Power Metallurgy Molding Machines Using Deep Learning Methods

Abstract

In this paper, an analysis and monitoring algorithm has been proposed for mold health evaluation using the vibration data. Two inertial measurement units (IMU, e.g., MPU6050) and an embedded system (NodeMCU) are implemented to collect vibration data on an MQTT protocol-based IoT(Internet of Things) platform from a powder metallurgy molding machine. In terms of data analysis, the vibration data on Z axis are segmented to label the upper/middle mold contact section and the corresponding vibration signal of sample friction on the X and Y axes are extracted. Using only historical vibration data from normal stamping, a BiLSTM+LSTM regression model is trained to predict normal stamping vibration data after several minutes later. Through the predicted stamping vibration data at the current time based on the historical data, the average mean square error (MSE) index compared with the observed data is calculated to evaluate the health status of the mold. Experimental results confirm that our prediction model training and evaluation indicators can notify operators in advance and an early warning system for mold damage is achieved.