An erecting hydraulic system fault diagnosis method, system and experimental platform

Abstract

This invention discloses a fault diagnosis method, system, and experimental platform for erecting hydraulic systems, relating to the field of hydraulic system fault diagnosis technology. Addressing the problems of high-load erecting hydraulic systems exhibiting strong fault concealment under complex working conditions, and the low efficiency and poor accuracy of traditional diagnostic methods, this invention utilizes an erecting hydraulic system fault simulation platform to conduct simulation experiments on erecting hydraulic system faults. Data from the simulation experiments is collected through a condition monitoring system, and the collected experimental data is preprocessed. A hybrid diagnostic model based on LSTM neural networks and KNN classifiers is used to perform real-time fault simulation diagnosis on the preprocessed experimental data, and the simulation diagnosis results are evaluated. This method effectively improves the automation level and accuracy of fault diagnosis, providing a reliable solution for the intelligent maintenance of erecting hydraulic systems.

Description

Technical Field

[0001] This invention relates to the field of hydraulic system fault diagnosis technology, and in particular to a method, system and experimental platform for diagnosing faults in a hydraulic system for erection. Background Technology

[0002] Erection hydraulic systems, as a critical hydraulic transmission and control system, are widely used in aerospace launch vehicles, heavy engineering machinery, and military equipment. Their main function is to provide stable hydraulic power to equipment, enabling precise erection, attitude adjustment, and load bearing of mechanical components. Due to the complex working environment of erection hydraulic systems, long-term operation can easily lead to various faults such as hydraulic cylinder leakage, pipeline blockage, and unstable system pressure, resulting in performance degradation or even functional failure. However, traditional fault diagnosis methods often rely on manual experience, which is inefficient and unable to address the hidden and coupled nature of erection hydraulic system faults, failing to meet the urgent demands of modern equipment for high real-time performance, high precision, and high reliability.

[0003] Therefore, researching fault simulation and intelligent diagnosis methods for erection hydraulic systems, and achieving accurate simulation of typical faults in erection hydraulic systems, synchronous acquisition of multi-dimensional data, and high-accuracy intelligent fault diagnosis, is of great significance for improving equipment reliability, reducing operation and maintenance costs, and ensuring mission success. Summary of the Invention

[0004] To address the aforementioned problems, this invention aims to provide a fault diagnosis method, system, and experimental platform for erecting hydraulic systems. By establishing a fault simulation platform for erecting hydraulic systems, it achieves accurate simulation of two typical fault types: pipeline blockage and hydraulic cylinder leakage, with each fault type categorized into three fault levels. Through a status monitoring system incorporating various types of sensors and data acquisition systems, it achieves synchronous acquisition of time-series data on pressure, flow, and displacement from multiple channels. Using an intelligent hybrid diagnostic model based on LSTM neural networks and KNN classifiers, it performs fault simulation diagnosis, successfully identifying six fault modes and one normal state with an accuracy rate of 82.8%. This effectively improves the automation level and accuracy of fault diagnosis, providing a reliable solution for the intelligent maintenance of erecting hydraulic systems.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: On one hand, the present invention provides a method for diagnosing faults in a hydraulic erection system, the method comprising: A fault simulation platform and condition monitoring system for vertical hydraulic systems were established. Based on the erection hydraulic system fault simulation platform, a simulation experiment of the erection hydraulic system fault was conducted. The data of the simulation experiment was collected by the condition monitoring system, and the experimental data collected by the condition monitoring system was preprocessed. Construct a hybrid diagnostic model based on LSTM neural network and KNN classifier, and train the hybrid diagnostic model; A hybrid diagnostic model is used to perform real-time fault simulation diagnosis on the preprocessed experimental data, and the simulation diagnosis results are evaluated.

[0006] Optionally, the working platform includes: tank; A hydraulic oil pump, connected to the oil tank, is used to provide power for the erection hydraulic system; Erection hydraulic cylinder; An electromagnetic proportional directional valve is connected to the hydraulic oil pump and the erecting hydraulic cylinder to control the inlet and outlet oil volume of the erecting hydraulic system. A proportional relief valve, connected to the oil tank and the electromagnetic proportional directional valve, is used to simulate the change of load force in the erection hydraulic system during the erection process. An electromagnetic relief valve is connected to the hydraulic pipeline between the hydraulic pump and the electromagnetic proportional directional valve, and is also connected to the oil tank, for limiting the system pressure of the hydraulic pipeline; Multiple throttle valves are respectively installed in the main oil circuit between the electromagnetic proportional directional valve and the erecting hydraulic cylinder, as well as inside the erecting hydraulic cylinder, to simulate pipeline blockage and hydraulic cylinder leakage.

[0007] Optionally, the pipeline blockage fault simulated by the throttle valve includes three fault states: mild pipeline blockage, moderate pipeline blockage, and severe pipeline blockage; the hydraulic cylinder leakage fault simulated by the throttle valve includes three fault states: mild hydraulic cylinder leakage, moderate hydraulic cylinder leakage, and severe hydraulic cylinder leakage.

[0008] Optionally, the condition monitoring system includes sensors and a data acquisition system. The sensors include displacement sensors, pressure sensors, and flow meters. All sensor signals are connected to the data acquisition system via signal lines and control lines.

[0009] Optionally, the specific steps for building the status monitoring system are as follows: Connect the displacement sensor, pressure sensor, and flow meter to the working platform; The program framework and parameters of the data acquisition system are built and configured. The signals collected by the displacement sensor, pressure sensor and flow meter are parameterized and calibrated using the data acquisition system. A PID closed-loop algorithm is used to control the piston rod displacement of the erecting hydraulic cylinder.

[0010] Optionally, the preprocessing includes format conversion and dataset partitioning of the collected experimental data.

[0011] Optionally, the specific process of constructing a hybrid diagnostic model based on LSTM neural networks and KNN classifiers includes: An LSTM neural network consisting of an input layer, an LSTM layer, a fully connected layer, and a SoftMAX layer is constructed to be responsible for feature extraction. The output of the LSTM neural network is used as the input of the KNN classifier to obtain a hybrid diagnostic model based on the LSTM neural network and the KNN classifier.

[0012] Optionally, the specific process of training the hybrid diagnostic model includes: Supervised training of the LSTM neural network was performed using the backpropagation algorithm and the Adam optimizer. Use a trained LSTM neural network to extract features from the training and test sets respectively; A KNN classifier is trained based on the extracted training set features and their labels. The extracted features were visualized using t-SNE technology to verify the feature quality and model training effect.

[0013] On the other hand, the present invention also provides a fault diagnosis system for an erection hydraulic system, comprising: The data acquisition and preprocessing module is used to acquire and preprocess experimental data under various fault simulation conditions; The diagnostic model building and training module is used to build and train a hybrid diagnostic model based on LSTM neural network and KNN classifier; The fault diagnosis and assessment module uses a hybrid diagnostic model to perform real-time fault simulation diagnosis and effect evaluation. Among them, the data acquisition and preprocessing module, the diagnostic model construction and training module, and the fault diagnosis and evaluation module are implemented based on the fault diagnosis method of the erection hydraulic system as described above.

[0014] Furthermore, the present invention also provides an experimental platform, which includes a simulation platform for simulating hydraulic system failures and a status monitoring system. The status monitoring system is embedded with the hydraulic system failure diagnosis system described above.

[0015] The beneficial effects of this invention are: (1) This invention establishes a working platform that can simulate two typical faults: pipeline blockage and hydraulic cylinder leakage. Each fault is divided into three levels: mild, moderate and severe. The degree of fault is precisely controlled by the valve opening of the throttle valve, simulating real working conditions of different fault levels. Faults with hidden characteristics are transformed into quantifiable parameter features, effectively reducing manpower and time costs. (2) This invention achieves the successful identification of 6 fault modes and 1 normal state by synchronously acquiring time-series data of pressure, flow and displacement of multiple channels through a state monitoring system containing multiple types of sensors and data acquisition systems, thereby realizing the automation and real-time nature of fault detection and identification and significantly improving the system maintenance efficiency. (3) This invention uses an intelligent hybrid diagnostic model based on LSTM neural network and KNN classifier to diagnose the degree of fault. Combining the feature extraction capability of deep learning with the efficiency of traditional classifier, the accuracy of identifying different fault degrees reaches 82.8%. It can identify potential faults in advance in the early stage of fault occurrence, avoid further deterioration of faults, effectively reduce the probability of sudden system failures, and provide protection for the safety of equipment and personnel. Attached Figure Description

[0016] Figure 1 This is a flowchart of the fault diagnosis method for the erection hydraulic system in this invention.

[0017] Figure 2 This is a schematic diagram of the erection hydraulic system in this invention.

[0018] Figure 3 This is a graph showing the relationship between the load force and the displacement of the hydraulic cylinder piston rod in this invention.

[0019] Figure 4 This is a hardware connection diagram for simulating pipeline blockage faults in this invention.

[0020] Figure 5 This is a hardware connection diagram for simulating internal leakage faults in a hydraulic cylinder in this invention.

[0021] Figure 6 This is a training error curve of the LSTM model in this invention.

[0022] Figure 7 This is a curve showing the test error of the LSTM model in this invention.

[0023] Figure 8 This is a graph showing the training accuracy of the LSTM model in this invention.

[0024] Figure 9 This is a graph showing the accuracy of the LSTM model tested in this invention.

[0025] Figure 10 This is a comparison chart of fault diagnosis results in this invention.

[0026] Figure 11 This is a fault classification confusion matrix diagram in this invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments, and not all of the embodiments. Example 1

[0028] Example 1 provides a method for diagnosing faults in a hydraulic erection system, as shown in the attached figure. Figure 1 As shown, the fault diagnosis method for the erection hydraulic system specifically includes the following steps: Step 1: Establish a fault simulation platform and condition monitoring system for the vertical hydraulic system; Optionally, step 1 includes the following sub-steps: Sub-step 101: Establish a working platform for simulating typical faults in the vertical hydraulic system; For details, see attached. Figure 2 As shown, a working platform for simulating typical faults of the erection hydraulic system is built based on the schematic diagram of the erection hydraulic system. The working platform includes an oil tank 1, a hydraulic oil pump 2, an electromagnetic proportional directional valve 4, a proportional relief valve 5, an electromagnetic relief valve 6, an erection hydraulic cylinder 3, and multiple throttle valves 7.

[0029] Hydraulic pump 2 is connected to oil tank 1 to provide power for the entire system. In this embodiment, hydraulic pump 2 is powered by 380V power and can reach a speed of 1730 rpm, which meets the power requirements of the experiment. The erecting hydraulic cylinder 3 is a two-stage telescopic hydraulic cylinder (hereinafter referred to as hydraulic cylinder), with a total stroke of 600mm. It includes two working chambers: a rodless chamber and a rod chamber. The inner diameter of the first-stage hydraulic cylinder is 32.20mm, and the inner diameter of the second-stage hydraulic cylinder is 62.90mm, which can meet the requirements of large-range displacement testing. The electromagnetic proportional directional valve 4 is connected to the hydraulic oil pump 2 and the erecting hydraulic cylinder 3, and is used to control the inflow and outflow of oil in the entire erecting hydraulic system, thereby realizing the system's operating condition switching function. Specifically, the electromagnetic proportional directional valve 4 is a three-position four-way electromagnetic proportional directional valve, with a control voltage of 24V DC voltage and an output signal of 4~20mA current signal. The electromagnetic proportional directional valve 4 mainly consists of a proportional electromagnet, a valve core, and a spring return mechanism. The proportional electromagnet receives electrical signals and outputs corresponding electromagnetic force, which drives the valve core to control the opening and closing of the oil circuit and the size of the opening. The spring return mechanism is used to reset the valve core by spring force when the electromagnetic force disappears, thereby restoring the system. When the electromagnetic proportional directional valve 4 is in the left position, oil enters the rodless chamber of the hydraulic cylinder, and the hydraulic cylinder performs the erecting action. When the electromagnetic proportional directional valve 4 is in the right position, oil enters the rod chamber, and the hydraulic cylinder retracts, thereby realizing the system's operating condition switching. The proportional relief valve 5 is connected to the oil tank 1 and the electromagnetic proportional directional valve 4, primarily functioning as a load-bearing device. By controlling the opening of the valve core of the proportional relief valve 5, the pressure in the reverse chamber of the hydraulic cylinder is controlled to simulate the change in load force during the erection process. The load force is the erection force exerted by the hydraulic cylinder to lift the load. Assuming the load mass is 10 tons, the relationship curve between the load force and the piston rod displacement of the hydraulic cylinder is shown in the attached figure. Figure 3 As shown; The electromagnetic relief valve 6 is connected to the hydraulic pipeline between the hydraulic oil pump 2 and the electromagnetic proportional directional valve 4, and is also connected to the oil tank 1. It is used to limit the system pressure of the hydraulic pipeline. When the system pressure exceeds the set pressure of the electromagnetic relief valve 6, the valve will automatically open to release excess hydraulic oil in the pipeline and send it back to the oil tank 1, preventing the system pressure from being too high and damaging the components or pipeline. Furthermore, the two main oil circuits connecting the hydraulic cylinder and the electromagnetic proportional directional valve 4 have a diameter of 10mm and a working pressure of 21MPa. On this basis, seven auxiliary test pipelines are added. All seven pipelines are made of high-pressure hoses with a diameter of 10mm, and conversion joints and tees of the same diameter are used at the pipeline connections to achieve interconnection of the pipelines.

[0030] A throttle valve 7 is installed in the main oil circuit from the electromagnetic proportional directional valve 4 to the inlet of the rodless chamber of the hydraulic cylinder. By adjusting different opening degrees of the throttle valve 7, different degrees of blockage in the main oil circuit can be simulated to achieve the simulation of mild, moderate, and severe hydraulic pipeline blockage faults. The hardware connection for pipeline blockage fault simulation is shown in the attached figure. Figure 4 As shown.

[0031] A bypass is connected in parallel between the rodless and rod chambers of the hydraulic cylinder. A throttle valve 7 is installed on this bypass. By controlling different opening degrees of the throttle valve 7, the flow area of ​​the bypass is changed, thereby simulating hydraulic cylinder internal leakage faults of varying severity caused by piston seal failure. The maximum leakage rate can reach 4 L / min. The hardware connection for simulating hydraulic cylinder internal leakage faults is shown in the attached figure. Figure 5 As shown.

[0032] In this embodiment, the throttle valve 7 is a straight-through high-pressure ball valve of model YGZQ-G3 / 8, with a pressure resistance of 31.5MPa.

[0033] Sub-step 102: Construct a status monitoring system for the erection hydraulic system, including sensors and a data acquisition system; Specifically, various types of high-precision sensors are configured for the acquisition of experimental data, including displacement sensors, pressure sensors, and flow meters (8). Among them, the displacement of the hydraulic cylinder piston rod is measured using a LEC-150 type wire-type displacement sensor, with a measurement range of 0~1500mm and a linear accuracy of ±0.05% FS; CS-PT300 piezoresistive pressure sensors are arranged at four key nodes of the hydraulic cylinder: the inlet, outlet, rodless chamber, and rod chamber. These sensors have a range of 0~40MPa and a comprehensive accuracy of 0.5 grade; the flow meter 8 is connected in series with the throttle valve 7, as shown in the attached diagram. Figure 2 As shown, flow meter 8 is an LWGY type intelligent turbine flow meter with a range of 3.3~20 L / min and an accuracy of ±1.0% of the reading. The specific parameters of flow meter 8 are shown in Table 1 below.

[0034] Table 1 Specific parameter configuration of flow meter 8 All sensor signals are connected to the data acquisition system via signal and control lines. The core of the data acquisition and control system utilizes the USB-3112 multifunction data acquisition card (hereinafter referred to as the acquisition card) from National Instruments (NI). This card is equipped with a 16-bit high-precision ADC, supports 8-channel differential analog input, and integrates a programmable signal conditioning module, automatically performing signal amplification, anti-aliasing filtering, and isolation processing. The measurement and control software is built on the LabVIEW 2024 development environment, with a modular program flowchart comprising four main functional modules: data acquisition, real-time display, process control, and data storage. The front panel features a human-machine interface that dynamically displays displacement-pressure-flow curves, allows setting control parameters, and records test data. The system uses a PID closed-loop control algorithm to accurately track the hydraulic cylinder's motion trajectory. Sampled data can be stored for later data processing and analysis. These sensors, the acquisition card, and the LabVIEW measurement and control software together form the condition monitoring system for the erection hydraulic system.

[0035] Furthermore, the program framework and parameters of the data acquisition card and LabVIEW measurement and control software are built and configured. First, the program framework for the data acquisition card to acquire voltage signals and output voltage signals is built to ensure that the acquired and output voltage signals are functioning correctly. Specifically, the "Open dev" function controls the acquisition card; the "Conv Out Source" function configures the clock source for the analog output task, determines the generation rate and synchronization method of the output signal, generates continuous waveforms (such as sine waves and square waves), and ensures the accuracy and real-time performance of the output signal; the "Trig Out Source" function configures the output behavior triggered in data acquisition or signal generation tasks; the "AI Chan Sel" function selects the analog input channel and configures the analog input channels to be read in the data acquisition task; the "Set Ai Range" function sets the signal range of the analog input channel to ensure that the data acquisition device correctly reads the amplitude range of the input signal; the "Ai Sample Mode" function defines the sampling mode and controls the continuity or finiteness of data acquisition; the "Ai Conct Type" function configures the terminal connection type of the analog input channel and determines the signal input method; the "Ai Sample Rate" function sets the sampling rate of the analog input (unit: Hz, the sampling period of this system is set to 10000ns); the "Ai Trig Source" function configures the trigger source of the analog input task, specifying the trigger signal source for starting or synchronizing acquisition; and the "Ai Conv..." function... The "Source" function sets the clock source for the analog input conversion and controls the timing of the analog-to-digital conversion. The "Aipretrig Points" function is used to configure the number of pre-trigger sampling points, i.e., the number of samples collected before the trigger event occurs. The "Cir Ai Fifo" function is used to buffer continuously collected data samples to avoid data loss. The "Ai SoftTrig" function is used to manually start or stop the acquisition task via software trigger.

[0036] Then, the displacement, pressure, and flow voltage signals collected by the displacement sensor, pressure sensor, and flow meter 8 are parametrically calibrated to ensure the accuracy of the sensor data. Data filtering and averaging are performed based on the number of samples and sampling rate. The summation of the data and the average value are then used as the current raw voltage value. The current control voltage is output through the AO output function, where the control voltage of the electromagnetic proportional directional valve 4 is 0~10V, and the control voltage of the proportional relief valve 5 is 0~2V. The raw voltage signals corresponding to the displacement, pressure, and flow collected by the sensors are converted into actual physical quantities using the formula node method. The conversion relationship for the displacement sensor is 4~20mA (0~10V) corresponding to 0~1600mm; the conversion relationship for the pressure sensor is 4~20mA (0~10V) corresponding to 0~25MPa; and the conversion relationship for the flow sensor is 4~20mA (0~10V) corresponding to 3.3~20 L / min.

[0037] Next, a PID closed-loop algorithm is used to control the displacement of the hydraulic cylinder piston rod, divided into a position closed-loop section, a load control section, and a manual control section. The position closed-loop section amplifies the set displacement and actual displacement by a factor of 100, sets the maximum and minimum values ​​of the PID controller output, and outputs control parameters from -100 to 100 through the PID function. This output is then divided by 20 to produce a voltage from -5V to 5V, used to control the electromagnetic proportional directional valve 4. When the difference between the actual displacement and the set displacement is greater than 0, the electromagnetic proportional directional valve 4 extends; when the difference is less than 0, it retracts; when the absolute value of the difference is less than 5mm, the electromagnetic proportional directional valve 4 is no longer controlled to prevent strong vibration and impact from continuous forward and reverse movements that could damage it. The load control section sets a gain based on the actual displacement of the hydraulic cylinder piston rod and outputs a corresponding voltage to control the valve opening of the proportional relief valve 5, thus achieving continuous loading control. The manual control section allows manual control of the electromagnetic proportional directional valve 4 and the proportional relief valve 5 via the LabVIEW panel buttons.

[0038] Finally, configure the front panel screenshot section, adjust the acquisition channel enable to the fully open state, adjust the acquisition voltage range to the maximum of 10.24V, adjust the sampling period to 10000ns, and configure the number of samples to 5000.

[0039] Step 2: Conduct simulation experiments on the failure of the erection hydraulic system based on the erection hydraulic system failure simulation platform, collect data from the simulation experiment through the condition monitoring system, and preprocess the experimental data collected by the condition monitoring system. Optionally, step 2 includes the following sub-steps: Sub-step 201: Conduct simulation experiments on the faults of the erection hydraulic system based on the erection hydraulic system fault simulation platform, and collect the data of the simulation experiment through the condition monitoring system; Specifically, a fault simulation experiment of the erection hydraulic system was conducted under load. The working platform was used to simulate two types of faults: pipeline blockage and hydraulic cylinder leakage. These two types of faults were further divided into mild, moderate, and severe fault states. Then, displacement, pressure, and flow data were collected under seven working conditions: normal state signal, mild pipeline blockage, moderate pipeline blockage, severe pipeline blockage, mild hydraulic cylinder leakage, moderate hydraulic cylinder leakage, and severe hydraulic cylinder leakage. 100 sets of data were collected for each working condition, and each set contained time-series data from 8 channels.

[0040] First, normal state signals are extracted. Under the premise of controlling all irrelevant variables, the condition monitoring system is used to measure the pressure in the rod and rodless chambers of the erection hydraulic system, the inlet and outlet pressures, the flow rates of the main and auxiliary hydraulic circuits, and the displacement data of the hydraulic cylinder piston rod. Signals are also collected from the inlet, outlet, rod and rodless chambers, and system pressure. By collecting pressure data from each port and flow data from the main and auxiliary hydraulic circuits, the pressure and flow performance of each part of the system during operation is comprehensively understood. These data are then compared and analyzed with various fault state data to verify whether the total pressure and flow signals contain sufficient characteristic information for fault simulation and diagnosis, and high-quality pressure and flow signal data are extracted.

[0041] Then, the throttle valve 7 installed on the main oil circuit was adjusted to put the main oil circuit into a state of slight pipeline blockage (blockage of about 5%), moderate pipeline blockage (blockage of about 10%), and severe pipeline blockage (blockage of about 15%). The throttle valves 7 on the other auxiliary oil circuits were in the normal closed state. The fault signals of slight pipeline blockage, moderate pipeline blockage, and severe pipeline blockage were extracted and compared with the signals of normal operation. It was found that there were significant differences in the amplitude, stability, and displacement speed of the pressure and flow signals. These differences provided an important basis for identifying the characteristics of individual pipeline blockage faults.

[0042] Furthermore, the leakage flow rate inside the hydraulic cylinder was set to 4% of the rated flow rate, corresponding to the initial wear stage of the simulated hydraulic cylinder piston rod. The throttle valve 7 on the auxiliary oil circuit was adjusted from a fully closed state to an open state, and the throttle valve 7 on the main oil circuit was adjusted to a fully open state (opening degree of 100%). The signal of slight leakage fault inside the hydraulic cylinder was extracted and compared with the signal of normal operation. It was found that the flow rate of the main oil circuit decreased from the original 14.12 L / min to 13.8 L / min, with a decrease rate parameter of about 4%, which is consistent with the set value of leakage flow rate inside the hydraulic cylinder and the difference is obvious, with an error of about 0.523 L / min; the inlet pressure decreased from 2.5 MPa under normal conditions to 1.5 MPa under the condition of slight leakage fault inside the hydraulic cylinder, with a significant decrease.

[0043] Furthermore, the leakage flow rate inside the hydraulic cylinder was set to 8% of the rated flow rate, corresponding to the mid-term wear stage of the simulated hydraulic cylinder piston rod. The throttle valve 7 on the auxiliary oil circuit was adjusted from a fully closed state to a semi-open state, while the throttle valve 7 on the main oil circuit remained fully open (100% opening). The signal of moderate leakage fault inside the hydraulic cylinder was extracted and compared with the signal under normal working conditions. It was found that the flow rate of the main oil circuit decreased from the original 14.12 L / min to 13.20 L / min, with a decrease rate parameter of about 8%, which is consistent with the set value of leakage flow rate inside the hydraulic cylinder and the difference is obvious, with an error of about 1.1302 L / min. The inlet pressure decreased from 2.5 MPa under normal conditions to 1.0 MPa under the moderate leakage fault condition inside the hydraulic cylinder, with a significant decrease.

[0044] Furthermore, the leakage flow rate inside the hydraulic cylinder was set to 12% of the rated flow rate, corresponding to the late wear stage of the simulated hydraulic cylinder piston rod. The throttle valve 7 on the auxiliary oil circuit was adjusted from a fully closed state to a semi-open state, while the throttle valve 7 on the main oil circuit remained fully open (100% opening). The severe leakage fault signal inside the hydraulic cylinder was extracted and compared with the signal under normal working conditions. It was found that the flow rate of the main oil circuit decreased from the original 14.12 L / min to 12.60 L / min, with a decrease rate parameter of about 12%, which is consistent with the set leakage flow rate value inside the hydraulic cylinder and the difference is obvious, with an error of about 1.69 L / min. The inlet pressure decreased from 2.5 MPa under normal conditions to 0.5 MPa under severe leakage fault conditions inside the hydraulic cylinder, with a significant decrease.

[0045] Sub-step 202: Convert the format of the experimental data collected by the condition monitoring system and divide the dataset; Specifically, the original 70*99 column data is processed and converted into 990*7 rows, with each 990 rows corresponding to one type of fault state, for a total of 7 types. After conversion, the data is imported into the MAT program for further subdivision, extracting 8 state parameters: main oil circuit flow rate, auxiliary oil circuit flow rate, inlet pressure, outlet pressure, rod chamber pressure, rodless chamber pressure, system pressure, and piston rod displacement. These parameters are then adjusted into a 6930*8 data set, completing the import of the original data.

[0046] Next, the dataset is divided in detail. For each node parameter, every 99 rows constitute one sample data, resulting in a total of 70 sample data. These are further divided into 7 categories according to 7 fault states, resulting in a total of 100 samples.

[0047] After partitioning, 10 data sources, from data_all_0 to data_all_9, are read. Each data source is divided into 7 categories, and each category contains 100 samples, resulting in 10*99=990 rows. Then, the data is reorganized by merging data from different data sources into similar categories and storing them in a specified position in the y{j} matrix for subsequent feature extraction. That is, the y variable is divided into 7 datasets, y1~y7, corresponding to the collected data of 7 different working conditions. Here, j represents the category. Each dataset contains 100 samples, and each sample contains a 9900*8 matrix, where 9900 represents the time step and 8 represents the number of sensors, i.e., 1 displacement sensor, 2 flow sensors, and 5 pressure sensors.

[0048] Finally, the dataset was divided into training and testing sets, with 80% used as the training set and 20% as the testing set. Specifically, each sample extracted from the dataset was divided into X (representing the input feature) and Y (label). Then, the format was converted, with X converted to cell format and each sample converted to an [8*99] format, which is time series data from 8 sensors.

[0049] Step 3: Construct a hybrid diagnostic model based on LSTM neural network and KNN classifier, and train the hybrid diagnostic model; Optionally, step 3 includes the following sub-steps: Sub-step 301: Construct a hybrid diagnostic model based on LSTM neural network and KNN classifier; Specifically, an LSTM neural network is constructed, comprising an input layer, an LSTM (Long Short-Term Memory) layer, a fully connected layer, and a Softmax layer. The input layer receives the feature data for each sample, with an input dimension of [8*99]. The LSTM layer captures the temporal dependencies in the data; in this embodiment, 64 hidden units represent the number of memory units in the LSTM layer. The fully connected layer maps the output of the LSTM layer to specific fault categories. The Softmax layer is used for pre-training the model, transforming the model's output into probability values ​​for each category.

[0050] In MATLAB, the structure of an LSTM layer is represented by three key gating modules: the input gate, the forget gate, and the output gate. Each module contains a sigmoid function layer and controls the transmission of bitwise multiplication information. Data flow regulation is achieved through a combination of gating weight matrices and hidden states, ultimately completing the modeling of long-series signals and thus enabling data manipulation and deep data processing. The specific process is as follows: First, there's the input gate, which contains a sigmoid activation function and an element-wise multiplication operation. It generates a vector of values ​​between 0 and 1 by introducing the input vector of the current time step and the hidden state vector of the previous time step into two different fully connected layers. Each element represents the degree to which information should be stored. Specifically, for a given element J representing the degree to which information will be stored in the cell state, all information generated in the current time step and the previous step is considered to be stored. Therefore, the input gate can be viewed as a selection gate that controls how the input at the current time step should affect the cell state.

[0051] The input gate determines how the cell state updates the input information at the current moment. Its calculation mainly involves two steps: first, the generation of the gating signal, which uses the sigmoid function to filter important features, and can be represented as... (1) In the formula, and For learnable parameters, This is the hidden state from the previous moment. This is the current input.

[0052] Second, the generation of candidate information, through the input gate... Weighted candidate input vector This ensures that only key features affect cell state. Candidate input vectors Generated by the tanh function, it can be represented as (2) In the formula, This is the weight matrix for this item. This is the bias matrix for this term.

[0053] Secondly, there's the forget gate. The LSTM forget gate uses a sigmoid function to generate a forgetting coefficient between 0 and 1, quantifying the priority of retaining historical information. 0 represents complete discarding, and 1 represents complete inheritance. Its core function is to enhance the network's ability to model long-range temporal associations by dynamically attenuating redundant historical features in cell states, which is crucial for improving the accuracy of sequence prediction tasks. The forget gate calculation formula is... (3) In the formula, For the output of the forget gate, Here is the weight matrix for the forget gate. The bias matrix of the forget gate; It can be used to train weight matrices or bias vectors; The function can be used to hide the state from the previous time step. With the current input Vector concatenation.

[0054] The parameters of the forget gate are adaptively adjusted based on the training data through the backpropagation algorithm, enabling the model to autonomously identify key patterns in the input sequence that need to be maintained in the long term.

[0055] Updating memory units is a crucial part of training LSTM neural networks; cell states Updated jointly by the forget gate and the input gate, it can be represented as: (4) In the formula, This is an element-wise multiplication operation. Controlling the decay of historical information, By integrating new information, dynamic maintenance of long-term memory can be achieved.

[0056] The output gate controls the flow direction of output data, and its core function is to determine the visibility of the memory cell state to the external network. The output gate filters valid information through a gating mechanism, ensuring that the network only outputs features with temporal correlation. The specific operation process of the output gate can be represented as follows: (5) In the formula, Here is the weight matrix of the output gate. Let be the bias matrix of the output gate. The output of the output gate has a value distributed in the range [0, 1], representing the degree of information accessibility in different dimensions.

[0057] After fusing the output gating signal with the memory cell state, the final output is formed, which can be represented as follows: (6) In the formula, the tanh function performs a nonlinear transformation on the cell state. The above calculation process realizes the dynamic coupling between the current memory state and the output gating signal, ensuring that the output contains effective information with temporal characteristics.

[0058] Through the calculation of gating mechanisms such as forget gate, input gate, memory unit, and output gate, the memory unit is eventually updated, thereby calculating the input data for each time step.

[0059] Furthermore, since KNN (K-Nearest Neighbors) is suitable for rapid prototype validation in small-scale data and low-dimensional spaces, it is used as the final classifier. The LSTM neural network is responsible for feature extraction, and KNN is responsible for final classification. The output of the last time step of the LSTM layer is used as the input feature of the KNN, ultimately realizing a hybrid diagnostic model of LSTM neural network and KNN classifier.

[0060] Sub-step 302: Train the hybrid diagnostic model and verify the model training effect; Specifically, training parameters are set, a preprocessed training dataset is used, and the LSTM neural network is trained in a supervised manner using the Adam optimizer to measure the difference between the model's predicted class and the true class, enabling the LSTM neural network to extract effective feature representations from the original time-series data. Key training parameters include a maximum of 100 training epochs, 16 training samples per batch, and ensuring that 20% of the data is used as the detection dataset during training to avoid overfitting. The parameter configuration of the LSTM neural network is shown in Table 2 below.

[0061] Table 2 Parameter configuration of LSTM neural network During model training, the LSTM model training error curve, test error curve, training accuracy curve, and test accuracy curve are monitored and plotted in real time, as shown in the attached figures. Figure 6 Appendix Figure 7 Appendix Figure 8 Appendix Figure 9 As shown, with the increase of training times, both training and testing errors steadily decreased and tended to converge, while the accuracy steadily increased and eventually stabilized at a high level. This indicates that the model training process is effective, without serious overfitting or underfitting, and the model has good learning ability.

[0062] The trained LSTM neural network is used to extract features from the training and test sets respectively. The specific steps are as follows: (1) Initialize the matrix for storing the features of the training set, with a size of [number of samples × 64], where 64 is the feature dimension; (2) Iterate through each training sample, convert the input data into a deep learning format dlarray (CBT = channel × batch × time step, i.e., the number of features is 8, the batch size is 1, and the time step is 99), take the output of the last time step of LSTM (feat(:, end)), and transpose it into a row vector and store it in the matrix. The dataset shape is [64×1]. Use this code to transpose it into a [1×64] row vector and store it in the feature matrix. (3) Initialize the matrix storing the features of the test set, which is in the same format as the training set, and perform the same feature extraction process on the test set as on the training set to ensure feature consistency.

[0063] The training set feature matrix extracted by the LSTM neural network and its corresponding fault labels are input into the KNN classifier for training, thus completing the final construction of the hybrid diagnostic model.

[0064] To verify the effectiveness of the features extracted by the LSTM neural network and the training effect of the hybrid model, t-SNE (t-distributed random nearest neighbor embedding) technology was used to perform dimensionality reduction and visualization analysis on the extracted high-dimensional features. Specifically, all feature vectors from the training and test sets were merged and uniformly reduced in dimensionality to avoid the bias in visualization results caused by data segmentation. The output dimension of t-SNE was set to 2, projecting the high-dimensional features onto a two-dimensional plane. By observing the aggregation and separation of sample points of different categories in the two-dimensional space, the quality of the features can be intuitively evaluated, providing preliminary evidence for the performance of subsequent fault simulation and diagnosis.

[0065] Step 4: Perform real-time fault simulation diagnosis on the preprocessed experimental data using a hybrid diagnostic model, and evaluate the simulation diagnosis results. Specifically, to diagnose single-sample fault data of the erection hydraulic system in real time, a user diagnostic interface is set up. This interface allows users to input a 99×8 sample data, and predicts its fault category through the following process: First, the sample data undergoes the same preprocessing as the training set. The preprocessed data is then input into the trained LSTM neural network. During training, the data format is [feature × time step × batch]. The output shape of FeatureNet is [64 × 1 × 99] (channel × batch × time step). Taking the last time step, the corresponding feat(:, end) represents extracting the feature set (shape 64 × 1) of all channels at the last time step, transposing it into a 1 × 64 row vector, adapting it to the KNN input format, and then using KNN prediction. The trained KNN model is then used to classify the 64-dimensional features, returning the category index to obtain the fault classification result.

[0066] Based on the experimental results, the final accuracy rate was calculated to be 82.8%.

[0067] The model's predictions are compared with the true labels, as shown in the attached figure. Figure 10 As shown in the figure, the red curve represents the actual fault category sequence, and the blue curve represents the category sequence predicted by the model. The higher the overlap between the two curves, the more accurate the model prediction. When the red curve and the blue curve overlap, it indicates that the prediction is accurate. The remaining blue scattered points are prediction error values, and the result meets the basic fault diagnosis requirements.

[0068] The confusion matrix for fault classification is attached. Figure 11As shown, this is a 7x7 grid matrix. The rows of the matrix represent the fault categories predicted by the model, and the columns represent the actual fault categories. The values ​​on the main diagonal of the confusion matrix represent the number of samples that were correctly classified, and the values ​​on the off-diagonal represent the number of samples that were misclassified. The figure clearly shows which type of fault the model misclassifies as another type of fault, further demonstrating that the fault classification results can meet the requirements of basic fault diagnosis. Example 2

[0069] Example 2 provides a fault diagnosis system for a lifting hydraulic system, including a data acquisition and preprocessing module, a diagnostic model construction and training module, and a fault diagnosis and evaluation module; The data acquisition and preprocessing module is used to acquire and preprocess experimental data under various fault simulation conditions; The diagnostic model building and training module is used to build and train a hybrid diagnostic model based on LSTM neural network and KNN classifier; The fault diagnosis and assessment module uses a hybrid diagnostic model to perform real-time fault simulation diagnosis and effect evaluation. The data acquisition and preprocessing module, the diagnostic model construction and training module, and the fault diagnosis and evaluation module are implemented based on the fault diagnosis method of the erection hydraulic system described in Example 1. Example 3

[0070] Example 3 provides an experimental platform, which includes a hydraulic system fault simulation platform for erection and a status monitoring system. The status monitoring system is embedded with a hydraulic system fault diagnosis system for erection as described in Example 2.

[0071] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing faults in a hydraulic erection system, characterized in that, The method includes: A fault simulation platform and condition monitoring system for vertical hydraulic systems were established. Based on the erection hydraulic system fault simulation platform, a simulation experiment of the erection hydraulic system fault was conducted. The data of the simulation experiment was collected by the condition monitoring system, and the experimental data collected by the condition monitoring system was preprocessed. Construct a hybrid diagnostic model based on LSTM neural network and KNN classifier, and train the hybrid diagnostic model; A hybrid diagnostic model is used to perform real-time fault simulation diagnosis on the preprocessed experimental data, and the simulation diagnosis results are evaluated.

2. The method for diagnosing faults in a hydraulic erection system according to claim 1, characterized in that, The working platform includes: Fuel tank (1); A hydraulic oil pump (2) is connected to the oil tank (1) and is used to provide power for the erection hydraulic system; Hydraulic cylinder for erection (3); The electromagnetic proportional directional valve (4) is connected to the hydraulic oil pump (2) and the erecting hydraulic cylinder (3) and is used to control the oil inflow and outflow of the erecting hydraulic system. A proportional relief valve (5) is connected to the oil tank (1) and the electromagnetic proportional directional valve (4) to simulate the change of load force in the erection hydraulic system during the erection process; The electromagnetic relief valve (6) is connected to the hydraulic pipeline between the hydraulic oil pump (2) and the electromagnetic proportional directional valve (4), and is also connected to the oil tank (1) to limit the system pressure of the hydraulic pipeline. Multiple throttle valves (7) are respectively installed in the main oil circuit between the electromagnetic proportional directional valve (4) and the erecting hydraulic cylinder (3) and inside the erecting hydraulic cylinder (3) to simulate pipeline blockage faults and hydraulic cylinder leakage faults.

3. The method for diagnosing faults in a hydraulic erection system according to claim 2, characterized in that: The throttle valve (7) simulates three types of pipeline blockage faults: mild pipeline blockage, moderate pipeline blockage, and severe pipeline blockage. The throttle valve (7) also simulates three types of hydraulic cylinder leakage faults: mild hydraulic cylinder leakage, moderate hydraulic cylinder leakage, and severe hydraulic cylinder leakage.

4. The method for diagnosing faults in a hydraulic erection system according to claim 3, characterized in that: The status monitoring system includes sensors and a data acquisition system. The sensors include displacement sensors, pressure sensors and flow meters (8). All sensor signals are connected to the data acquisition system through signal lines and control lines.

5. The method for diagnosing faults in a hydraulic erection system according to claim 4, characterized in that, The specific steps for building the status monitoring system are as follows: Connect the displacement sensor, pressure sensor and flow meter (8) to the working platform; The program framework and parameters of the data acquisition system are built and configured. The signals collected by the displacement sensor, pressure sensor and flow meter (8) are parameterized and calibrated by the data acquisition system. The displacement of the piston rod of the erecting hydraulic cylinder (3) is controlled by a PID closed-loop algorithm.

6. The method for diagnosing faults in a hydraulic erection system according to claim 5, characterized in that: The preprocessing includes format conversion and dataset partitioning of the collected experimental data.

7. A method for diagnosing faults in a hydraulic lifting system according to claim 6, characterized in that, The specific process of constructing a hybrid diagnostic model based on LSTM neural network and KNN classifier includes: An LSTM neural network consisting of an input layer, an LSTM layer, a fully connected layer, and a SoftMAX layer is constructed to be responsible for feature extraction. The output of the LSTM neural network is used as the input of the KNN classifier to obtain a hybrid diagnostic model based on the LSTM neural network and the KNN classifier.

8. A method for diagnosing faults in a hydraulic lifting system according to claim 7, characterized in that, The specific process of training the hybrid diagnostic model includes: Supervised training of the LSTM neural network was performed using the backpropagation algorithm and the Adam optimizer. Use a trained LSTM neural network to extract features from the training and test sets respectively; A KNN classifier is trained based on the extracted training set features and their labels. The extracted features were visualized using t-SNE technology to verify the feature quality and model training effect.

9. A fault diagnosis system for a hydraulic erection system, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess experimental data under various fault simulation conditions; The diagnostic model building and training module is used to build and train a hybrid diagnostic model based on LSTM neural network and KNN classifier; The fault diagnosis and assessment module uses a hybrid diagnostic model to perform real-time fault simulation diagnosis and effect evaluation. The data acquisition and preprocessing module, the diagnostic model construction and training module, and the fault diagnosis and evaluation module are implemented based on the fault diagnosis method of the erection hydraulic system as described in any one of claims 1-8.

10. An experimental platform, characterized in that, The experimental platform includes a fault simulation platform for the erection hydraulic system and a status monitoring system, wherein the status monitoring system is embedded with the fault diagnosis system for the erection hydraulic system as described in claim 9.

Images