Research on fault warning technology for cutting part of cantilever roadheader based on virtual and real fusion data

Currently, the fault warning technology for the cutting part of cantilever roadheader relies on traditional data collection methods. In the operation process of the cutting part of the roadheader, problems such as difficulty in obtaining signals and high noise limit the capability to predict and warn faults in the cutting part of the roadheader. In order to solve the above problems, a fault warning method for the cutting part of cantilever roadheader based on virtual and real fusion data is proposed. The method performs three-dimensional solid modeling of the cutting section of a cantilever roadheader. It uses the automatic dynamic analysis of mechanical systems (ADAMS) to obtain virtual data of the cutting section's mechanical system, constructs its dynamic simulation model to obtain virtual data. The method uses the cosine similarity function to characterize its similarity with real data to verify the credibility of the virtual data. The method uses Bayesian estimation and adaptive complementary weighted fusion methods to perform similarity association and complementary association fusion on virtual and real data, respectively, to obtain virtual and real fusion data. In response to the problem that the learning efficiency of traditional self-organizing mapping (SOM) neural networks is easily affected by the learning rate, a fault warning model based on an improved SOM neural network is established. A monotonic decreasing function about time is introduced to train the SOM neural network, ensuring both the learning rate and the stability of the model. The method inputs the fused data into the fault warning model based on SOM neural network to determine the winning neuron and adjust its weight. The method calculates the distance between the real data and the winning neuron and adjusts its weight to achieve fault warning. The experimental results show that the average operating efficiency of the improved SOM neural network can be improved by 35.84%. The fault warning method for the cutting part of a cantilever roadheader based on virtual and real fusion data can successfully predict the types of single and composite faults, with a prediction accuracy of 83.33%.