Fault diagnosis system for puffing machine with Bayesian-optimized convolutional neural network and multi-head attention based on multi-source fusion

Due to its high-temperature and high-pressure operating environment, food/feed puffing machines are prone to faults such as cavity blockage and cutter wear. This paper presents the design of a fault diagnosis system for puffing machines (food/feed processing equipment that expands or puffs agricultural products), based on a convolutional neural network and a multi-head attention mechanism model, which incorporates Bayesian optimization. The system combines multi-source information fusion, capturing patterns and characteristics associated with fault states by monitoring various sources of information, such as temperature, noise signals, main motor current and vibration signals from key components. Hyperparameters are optimized through Bayesian optimization to obtain the optimal parameter model. The integration of convolutional neural networks and multi-head attention mechanisms enables the simultaneous capture of both local and global information, thereby enhancing data comprehension. Experimental results demonstrate that the system successfully diagnoses puffing machine faults, achieving an average recognition accuracy of 98.8% across various operating conditions, highlighting its high accuracy, generalization ability and robustness.