Health Assessment For Piston Pump With Multi-sensor Information Fusion And Deep Forest

The piston pump is an important power component of the hydraulic system and is widely used in high pressure and high flow situations,such as engineering machinery and ships.Due to the characteristics of high pressure and high flow of the piston pump,the performance of the piston pump will gradually decline with the increase of working time.The main reasons are wear,friction,shock,vibration,high temperature,corrosion,etc.The decline of the performance of the piston pump will affect the working state of the hydraulic system,which may cause system failure or even major safety accidents in serious cases.Therefore,it is particularly important to perform accurate health assessment of the piston pump.The traditional mechanism analysis can analyze the cause of the decline of the health state by analyzing the microstructure,fluid characteristics,and mechanical characteristics of the piston pump.The working state of the piston pump can be obtained through related performance test analysis.However,these methods often require disassembly testing of the piston pump,which is difficult to perform during the work of the piston pump.The data-driven approach has achieved good results in piston pump health assessment and fault diagnosis.The method for assessing the health of a piston pump is to analyze the collected piston pump operating data and establish a corresponding mathematical model.The performance degradation of the piston pump is often reflected in multiple semaphores.Therefore,the method of multi-sensor information fusion can more accurately evaluate the health status of the piston pump.Under actual working conditions,sometimes only the pressure signal of the hydraulic system can be collected,and sometimes multiple sensor signals can be extracted.Two different piston pump health assessments are proposed for two different states of the piston pump.(1)Perform a health assessment when the piston pump can only provide a single pressure signal.This paper proposes a piston pump health assessment method using empirical mode decomposition and deep forest.The empirical mode decomposition is used to decompose the original signal and extract feature vectors.The deep forest algorithm is used to evaluate the health status.Finally,the experimental results are compared with traditional machine learning algorithms.Experimental results show that the proposed method can establish an effective health assessment model under the single pressure signal condition.(2)Perform a health assessment when the piston pump can provide multiple types of sensor signals.The feature-level multi-sensor information fusion method is used to first calculate the characteristic signals of each sensor and use the improved feature fusion method for feature fusion.Finally,the deep forest is used to evaluate the health status of the piston pump,and the stability of the model is compared under different training set proportions.The test results show that the method improves the accuracy of the health assessment compared to the single-signal condition,and it can be used to build a health assessment model for piston pump with multiple sensor signals.By comparing the analysis of two different signal situations,it can be seen that the proposed method can effectively evaluate the health status when the actual working condition has only a single pressure signal.When the actual working conditions use multiple types of sensor signals,the results of the health assessment are more comprehensive and accurate.Based on the above-mentioned proposed method of piston pump health assessment,a piston pump health assessment system is designed.The software system function design and database design are completed,and the system software development is finally completed.The system can realize the functions of raw data management,health assessment model management and display of prediction results,and achieve the purpose of health assessment of the piston pump.