Research On The Monitoring And Diagnosis Method Of Hydraulic System Operation Status Based On Functional Data Analysi

Hydraulic system is a core component of national major technical equipment and its reliable and stable operation is crucial to ensuring high-quality and safe service.However,due to the complexity of the hydraulic system circuits,it is susceptible to periodic load changes and random factors such as component leakage,overheating,vibration,and noise during operation.As a result,the system often experiences problems such as delayed out-ofcontrol alarms and abnormal operating states.Existing mainstream process monitoring and diagnostic methods based on statistical models such as principal component analysis have not fully integrated the characteristics of hydraulic system data,such as different sampling rates,complex coupling relationships between multi-cycle and multi-channel variables,and dispersed failure distributions,making it difficult to uncover the functional data characteristics of data samples and ensure the efficient and reliable operation of the system.Therefore,treating hydraulic system operation data as a functional process for analysis,exploring the evolutionary laws of functional data,identifying key influencing factors,constructing functional state space models,and implementing system operation state monitoring and diagnosis has significant theoretical and practical value for ensuring reliable and stable operation of hydraulic systems.According to the key data characteristics of hydraulic system,a functional K-means method is used to analyze the operational data of the hydraulic system,achieving efficient data representation and clustering.Then a functional state space model is proposed for data modeling.Its one-step prediction errors were computed,and the exponentially weighted moving average(EWMA)statistic was obtained for further online monitoring and real time anomaly detection of the hydraulic system.Last,a fault diagnosis model based on IMPACat Boost was constructed,which facilitated the diagnosis of abnormal operational status of the system.The proposed research framework was applied and validated in a real hydraulic system.The main contributions of this paper can be summarized as follows:(1)Clustering analysis for operational data of hydraulic system based on functional expression.To describe the operational data with high dimensionality and periodic changes,a functional decomposition approach using Fourier analysis for data smoothing was applied.Then K-Means clustering method was utilized for clustering.Based on the clustering results,further functional data modeling for each cluster separately can be further conducted.(2)By designing a function-based state space model,the operating status of the hydraulic system can be modeled.To address the complex coupling relationships between different stages and sensors,including inter-stage correlations,inter-channel correlations,and self-regressive properties of profile data,as well as differences between operating stages and sensors,a function basis expansion method is conducted to describe within-profile and cross-profile correlations between different variables.Then EM algorithm and Kalman filtering are applied for model inference.(3)Monitoring of operating status for hydraulic system based on function-based state space models.Based on the proposed function-based state-space model,a monitoring statistic is conducted by using the one-step prediction error.Furthermore,a group index weighted moving average control chart is established to effectively reduce the alarm delay of abnormal states.Numerical study and practical case analysis indicate that the chart exhibits better stability,and shows more robust performance in the identification of operating conditions of the hydraulic system than methods in the literature.(4)Anomaly diagnosis of the operating status of hydraulic system based on ensemble learning.To diagnose faults in the hydraulic system,an improved marine predator algorithm based on nonlinear convergence factors and fused rank-position update strategies is proposed.Combining the Cat Boost feature importance ranking with the improved marine predator feature selection method,the optimal feature subset related to the fault is obtained,and an IMPA-Cat Boost diagnostic model for the abnormal state of the hydraulic system is constructed.The experimental results demonstrate that the IMPA-Cat Boost algorithm exhibits high accuracy and generalization ability in diagnosing the fault types of various components.(5)Application of function data analysis in monitoring and diagnosing the state of hydraulic system.Based on the research findings above,the modeling and monitoring methods based on functional space models were applied to the operation monitoring of an real pneumatic experimental platform,and the abnormal state diagnosis method based on ensemble learning was used to identify and diagnose abnormal state types during the operation of the experimental platform.The experiment shows that the proposed methods can accurately model the operational states,issue alarms for out-of-control states,and identify fault types,which verifies the effectiveness of the proposed methods.