Hydraulic System Fault Diagnosis Based On Multi-source Information Ensemble

The hydraulic system is the core device of modern mechanical equipment.It is widely used in agricultural production,transportation,industrial manufacturing,and other fields.Due to their long-term harsh environment and complex working conditions,individual components are more difficult to observe and monitor visually than other mechanical equipment,which results in hydraulic system failures having unique characteristics and rules.Compared to other mechanical faults,hydraulic system faults are more difficult to locate,they occur randomly and the failure modes are more diverse.Therefore,timely and accurate identification of the types of faults in hydraulic systems can help to reduce operating costs and enhance system stability and reliability.However,there are still many problems in hydraulic system fault diagnosis that need to be solved:firstly,most of the research relies too much on vibration analysis and cannot make full use of the multi-physical parameter information in the system to ensure the accuracy of diagnosis.Secondly,a great deal of current research in hydraulic systems is limited to dealing with data from individual sensors,which may not yield good diagnostic performance for complex systems with multi-sensor data.To address these issues,the research in this paper is as follows:(1)A research on fault diagnosis of oil-hydraulic systems based on convolutional neural networks with multiple sources of information.Deep learning can be used for feature extraction and feature selection from raw data and for classification,regression and prediction tasks by means of automatic learning.Convolutional neural networks,as one of the most effective algorithms in deep learning,can effectively avoid information loss caused during data pre-processing and ensure the authenticity of the data,as well as construct input-output relationships in complex systems with deep networks.Therefore,in this paper,the four physical and chemical indicators of fluid,namely viscosity,moisture,acid value and contamination,are directly used as inputs to the convolutional neural network to fully reflect the state information of the fluid and output the probability of "normal" or "fault" category,successfully constructing a mapping model between fluid data and hydraulic The proposed model is compared with BP neural network and support vector machine diagnostic model in terms of fault diagnosis accuracy,and the diagnostic accuracy of the proposed method is higher than other methods through experimental verification,with an average diagnostic accuracy of 94.3%.(2)A research of multi-class fault diagnosis in multi-source information hydraulic systems based on Blending ensemble learning.Ensemble learning integrates weak and independent models together with a combinatorial strategy to obtain more accurate results than individual models.Blending method as an effective combinatorial strategy in ensemble learning can use basic models with different types,structures,and parameter settings to obtain diverse model combinations and improve the system’s adaptability to states and generalization ability.Therefore,this paper pretrains the convolutional neural network models using multiple sources of information to obtain multiple pretrained convolutional neural network models,and integrates the pre-trained convolutional neural network models together with the Blending integrated learning method based on random forest meta-learner to obtain a multi-class fault diagnosis model for hydraulic systems to complete the diagnosis of multi-class faults in hydraulic systems.The experimental results show that the model achieves 96.9% accuracy in the diagnosis of multi-class faults in hydraulic systems by validating the data measured in the hydraulic system experimental bench.