Research On Gearbox Health Prediction Based On Multi-scale PSO-BP Neural Network

Gear box has the advantage of a wide range of applications,is widely used in chemical industry,petroleum,wind power and other fields,due to complex working environment and long-term in high load work and by violent vibration and shock,in the case of improper maintenance,easy to break down,and even cause malignant accidents,as well as significant economic losses.In order to avoid accidents and ensure the healthy and stable operation of gear box,oil monitoring technology is used to monitor the physical and chemical wear of oil and metal,analyze the causes,and take timely measures to prevent or solve the fault.However,there are miscellaneous information in the oil,so it is difficult to evaluate the health state of the gearbox only by expert experience and fuzzy comprehensive evaluation model.BP neural network is widely used in the field of mechanical health prediction due to its strong adaptive ability and generalization ability,but it is easy to fall into the local optimal solution.Therefore,BP neural network is optimized by combining particle swarm optimization algorithm(PSO)with global optimization ability and EMD algorithm with multi-scale information extraction ability.A gear box health parameter model based on multi-scale PSO-BP prediction was constructed,and the trend prediction function with objectivity and pertinent advantages was combined to predict the gear box health state.(1)Grey correlation analysis was used to determine the main oil parameters affecting the health state of the gear box.In this paper,data of oil wear and additive elements were collected by mass spectrometer,and moisture,acid value,oxidation value and viscosity were obtained by infrared oil detector and viscometer.Due to the existence of irrelevant data in the oil,which affects the health assessment results,in order to eliminate redundant information,the initial value method of grey correlation analysis is adopted to conduct dimensionless processing on the oil data,so as to eliminate the problem that the analysis accuracy decreases due to the large difference between the data.Then,the difference sequence calculation of the data is carried out,and the correlation coefficient matrix of the oil sample is obtained.Then the correlation coefficient of each oil component is averaged,so as to determine the correlation degree of each oil component.The main wear elements,additives and physical and chemical parameters that affect the operation of the gear box are selected through the grey correlation degree,and they are taken as the health parameters of the gear box.(2)Aiming at the problem that BP neural network is easy to fall into the local optimal solution when fault diagnosis is carried out,the BP model is optimized by combining EMD and particle swarm optimization algorithm,and the health parameter model based on multi-scale PSO-BP prediction is established to improve the classification accuracy and obtain the sample residual.The selected health parameter information was placed at the input end of BP neural network,and the set digit "1" of gear normal operation condition was placed at the output end of the neural network.The BP neural network was optimized by using the particle swarm optimization algorithm for population position,fitness,speed and acceleration,and then EMD algorithm was used to establish a multi-scale information extraction channel at the input port of BP.Based on this,a gearbox health state parameter model based on multi-scale PSO-BP prediction was constructed.The results show that compared with BP prediction model and PSO-BP prediction model,the health state parameter model has smaller residual error and higher prediction accuracy.The sample residuals were obtained by comparing the experimental values with the predicted values of the model,so as to prepare for predicting the health state of the gearbox.(3)Aiming at the low accuracy of gear box health status assessment with fuzzy comprehensive evaluation model,the health index was proposed.The sample residual was combined with the trend prediction function to build the gear box health state evaluation model based on trend prediction.The sample residual was analyzed by the trend prediction function,and the gear box health state was evaluated by combining the health degree,which was divided into five health states: excellent,good,average,poor and poor.Finally,the reliability of the model is verified by pollution degree analysis and ferrographic image.The results show that the health status evaluation model based on trend prediction has higher accuracy than the fuzzy comprehensive evaluation model,and the classification accuracy is increased by 40%.The model provides a reference for the health evaluation of similar mechanical equipment.