An Early Fault Diagnosis Method and Device for the Main Pump of a Hydraulic Shovel Based on an LSTM Flow Prediction Model

Abstract

The present invention discloses a method and device for early fault diagnosis of the main pump of a hydraulic shovel based on an LSTM flow prediction model. The method includes: collecting data at the edge end and screening the data related to the main pump as input data; performing neatness processing on the collected input data; splitting the processed input data into training and testing sets; training the LSTM model, using gridsearchCV to obtain the optimal parameters of the model, and saving the optimal model; using the saved LSTM model to predict the flow rate of the main pump. Through the good performance of the LSTM model in a long time series, the present invention applies it to mining excavation. Under the condition of uniform right rotation, other multi-time series data that can be actually measured are used to predict the main pump flow rate data that is not easy to actually measure. The predicted main pump flow rate is relatively accurate and the error is small. It can be considered that this model can effectively predict the main pump flow rate.

Description

Technical Field

[0001] The present invention belongs to the field of industrial fault diagnosis, and particularly relates to a method and device for early fault diagnosis of the main pump of a hydraulic shovel based on an LSTM flow prediction model. Background Art

[0002] As one of the important power components of mechanical equipment, the hydraulic pump is an important part of the hydraulic system. Its main function is to convert the mechanical energy input by the prime mover into hydraulic energy. It is widely used in modern industrial equipment such as mining and metallurgy, national defense construction, power systems, agricultural machinery, chemical engineering and petroleum, and shipbuilding. However, due to the complex structure of the hydraulic pump itself, it is difficult to actually measure the flow rate of its main pump. Therefore, studying the method of predicting the main pump flow rate, monitoring the state of the hydraulic pump, discovering the early faults of the hydraulic pump in advance, and taking maintenance measures in time have important practical significance and application value for reducing casualties, reducing economic losses, and ensuring the normal operation of mechanical equipment.

[0003] In recent years, with the continuous development of deep learning technology, some deep learning models have gradually been applied to the research of time series data. The deep learning model is a deep neural network model with multiple non-linear mapping layers, which can abstract and extract features layer by layer from the input signal and discover deeper potential laws. Among many deep learning models, the Recurrent Neural Network (RNN) introduces the concept of time series into the network structure design, making it show stronger adaptability in time series data analysis. Among many variants of RNN, the Long Short-Term Memory (LSTM) model makes up for the problems of gradient disappearance and gradient explosion of RNN, as well as insufficient long-term memory ability, enabling the recurrent neural network to effectively utilize long-distance time series information. There have been many successful application cases of the LSTM model in the research of time series data in different fields, including language modeling related to text language, speech recognition, machine translation, audio and video data analysis related to multimedia, picture caption modeling, traffic flow velocity prediction related to road transportation, and protein secondary structure sequence prediction related to medicine.

[0004] Based on the research and learning of the above LSTM model, the present invention proposes a method for predicting the main pump flow rate using the LSTM model to solve the problem that it is difficult to actually measure the flow rate of the main pump. Summary of the Invention

[0005] In order to solve the problem of difficult measurement of the main pump flow rate, the present invention proposes a method and device for early fault diagnosis of the main pump of a hydraulic shovel based on an LSTM flow prediction model.

[0006] The solutions adopted by the present invention are as follows:

[0007] The present invention provides a method for early fault diagnosis of the main pump of a hydraulic shovel based on an LSTM flow prediction model. The method is based on the LSTM model for prediction and includes:

[0008] Collect data at the edge side and screen the data related to the main pump as input data;

[0009] Perform neatness processing on the collected input data;

[0010] Split the processed input data into training and testing sets;

[0011] Train the LSTM model, use gridsearchCV to obtain the optimal parameters of the model, and save the optimal model;

[0012] Use the saved LSTM model to predict the flow rate of the main pump.

[0013] In one implementation, the step of collecting data at the edge side and screening the data related to the main pump as input data includes:

[0014] Divide the working conditions by the pilot pressure values of various actions to obtain the data under uniform rotation;

[0015] Screen the multivariate data related to the main pump; the multivariate time series data includes: the pressure of main pump 2, the engine speed, the engine torque percentage, the BP19 - right swing pilot pressure, and the Y12 - right swing proportional valve current.

[0016] In one implementation, the step of performing neatness processing on the collected input data includes: performing interpolation processing on the low - frequency data to make its frequency consistent with the high - frequency data, and aligning the processed data in time.

[0017] In one implementation, the step of splitting the processed input data into training and testing sets includes:

[0018] Split the input data into training and testing sets according to the time series;

[0019] Use MinMaxScaler to scale the data;

[0020] Predict the 31st target value through the input values of the past 30 time points (all features except the target column), and use the sliding window method to slide down sequentially for prediction.

[0021] Predict the 31st target value through the input values of the past 30 time points (all features except the target column), and use the sliding window method to slide down sequentially for prediction.

[0022] In one embodiment, when training the LSTM model, if the LSTM model makes more hyperparameter adjustments, add more layers; if the dataset is very large, increase the number of epochs and units in the LSTM model.

[0023] In one embodiment, when predicting the main pump flow rate using the saved LSTM model, the data predicted by the model needs to be inversely scaled to obtain the true predicted flow rate data.

[0024] In one embodiment, after predicting the main pump flow rate using the saved LSTM model, it further includes: calculating the volumetric efficiency using the predicted main pump flow rate to evaluate whether the predicted flow rate is accurate.

[0025] In one embodiment, under the condition of uniform right rotation, the volumetric efficiency = actual flow rate / theoretical flow rate; where the actual flow rate = LSTM predicted flow rate, and the theoretical flow rate = rated displacement of the pump × 0.001 (unit conversion) × measured engine speed.

[0026] In one embodiment, the method for evaluating whether the predicted flow rate is accurate is: determining whether the volumetric efficiency of the main pump is within the normal range of 0.95 - 0.98; if so, it proves that the predicted flow rate is accurate.

[0027] The present invention also provides a hydraulic shovel main pump early fault diagnosis device based on an LSTM flow prediction model, and this device includes:

[0028] An acquisition module, used to acquire data at the edge and screen the data related to the main pump as input data;

[0029] A data processing module, used to process the acquired input data for neatness;

[0030] A test splitting module, used to split the processed input data into training and testing data;

[0031] A training module, used to train the LSTM model, obtain the optimal parameters of the model using girdsearchCV, and save the optimal model;

[0032] A prediction module, used to predict the main pump flow rate using the saved LSTM model.

[0033] Advantages of the present invention

[0034] Due to the good performance of the LSTM model in a long time series, the present invention applies it to mining excavation. Under the condition of uniform right rotation, other actually measurable multi - time - series data are used to predict the main pump flow rate data that cannot be actually measured. The predicted main pump flow rate is relatively accurate with small errors. Description of the drawings

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art.

[0036] Figure 1 It is the computational graph structure of the deep LSTM neural network;

[0037] Figure 2 It is the flowchart of a method for early fault diagnosis of the main pump of a hydraulic shovel based on the LSTM flow prediction model provided by an embodiment of the present invention;

[0038] Figure 3 It is the flowchart of a method for early fault diagnosis of the main pump of a hydraulic shovel based on the LSTM flow prediction model provided by the optimal embodiment of the present invention;

[0039] Figure 4 It is the schematic structural diagram of a device for early fault diagnosis of the main pump of a hydraulic shovel based on the LSTM flow prediction model provided by an embodiment of the present invention;

[0040] Figure 5 It is the comparison diagram of the predicted flow rate of the main pump 2 and the actual flow rate by this method in a specific embodiment;

[0041] Figure 6 It is the volumetric efficiency diagram of the predicted flow rate of the main pump 2 by this method in a specific embodiment. Embodiment

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0043] In actual working conditions, it is difficult to actually measure the flow rate of the main pump. However, we can predict the flow rate of the main pump through other measured multivariate time series data using a deep neural network model. Therefore, we attempt to apply the LSTM model to the industrial field to predict data that cannot be actually measured with other measurable multivariate time series data in the LSTM model. For the target column data to be predicted in this article, a prediction method based on the LSTM recurrent neural network is proposed, including the detailed design of a three-layer (input layer, hidden layer, and output layer) network structure, as Figure 1 shown, as well as the implementation algorithms for network training and network prediction. On this basis, with the goal of minimizing the prediction error, an optimal selection algorithm for the parameters of the LSTM prediction model based on grid search is further proposed. Experiments are carried out using the measured data under the condition of uniform right rotation. The experimental results demonstrate the excellent performance of the proposed LSTM prediction model and its parameter optimal selection algorithm in time series prediction.

[0044] As Figure 2 shown, the present invention proposes a method for early fault diagnosis of the main pump of a hydraulic shovel based on an LSTM flow prediction model, and the method includes the following steps:

[0045] Step S100: Collect data at the edge end and screen the data related to the main pump as input data;

[0046] In the embodiment of the present application, collecting data at the edge end and screening the data related to the main pump as input data specifically includes:

[0047] Dividing the working conditions by the pilot pressure values of various actions to obtain the data under uniform slewing;

[0048] Screening the multi-source data related to the main pump; the multi-source time series data includes: the pressure of main pump 2, engine speed, engine torque percentage, BP19 - right slewing pilot pressure, and Y12 - right slewing proportional valve current.

[0049] Specifically, first collect data at the edge end and divide the working conditions by the pilot pressure values of various actions. When an action occurs, the corresponding pilot pressure will generate a corresponding value, and the other pilot pressures will not change. For example, when slewing to the right at a constant speed, the right slewing pilot pressure is stable at about 41, and the other pilot pressures basically tend to 0, so as to obtain the data under right slewing at a constant speed. Then screen the multi-source time series data related to the main pump, that is, the pressure of main pump 2, engine speed, engine torque percentage, BP19 - right slewing pilot pressure, and Y12 - right slewing proportional valve current.

[0050] Step S200: Process the collected input data for neatness;

[0051] In the embodiment of the present application, processing the collected input data for neatness includes: performing interpolation processing on the low-frequency data to make its frequency consistent with the high-frequency data, and aligning the processed data in time.

[0052] Specifically, in this embodiment, the sampling frequency of the main pump flow to be predicted and the pressure of main pump 2 collected is 200hz, the sampling frequency of engine data is 20hz, and the sampling frequency of BP19 - right slewing pilot pressure and Y12 - right slewing proportional valve current is 5hz. Perform interpolation processing on the low-frequency data to make its frequency consistent with the high-frequency data. Align the processed data in time.

[0053] Step S300: Split the processed input data into training and testing sets;

[0054] In the embodiment of the present application, splitting the processed input data into training and testing sets includes:

[0055] Split the input data into training and test sets according to the time series;

[0056] Scale the data using MinMaxScaler;

[0057] Predict the 31st target value based on the input values of the past 30 time points (all features excluding the target column), and use the sliding window method to slide down sequentially for prediction.

[0058] Specifically, split the input data into training and test sets. Note that the data cannot be shuffled here because it must be sequential in the time series. Since the data range is very large and they are not scaled within the same range, in order to avoid prediction errors, let's first scale the data using MinMaxScaler. Use the input values of the past 30 time points (all features excluding the target column) in the LSTM model to predict the 31st target value. Use the sliding window method to slide down sequentially for prediction.

[0059] Step S400: Train the LSTM model, use girdsearchCV to obtain the best parameters of the model, and save the best model;

[0060] Optionally, when training the LSTM model, if the LSTM model makes more hyperparameter adjustments, add more layers; if the dataset is very large, increase the epochs and units in the LSTM model.

[0061] Specifically, when training the model, use girdsearchCV to perform some hyperparameter adjustments to find the basic model. If the model makes more hyperparameter adjustments, more layers can also be added. However, if the dataset is very large, it is recommended to increase the epochs and units in the LSTM model. Check the best parameters of the model and save the best model.

[0062] Step S500: Predict the main pump flow rate using the saved LSTM model.

[0063] In the embodiments of the present application, when predicting the main pump flow rate using the saved LSTM model based on the implementation of step S300, it is necessary to inverse-scale the data predicted by the model to obtain the real predicted flow rate data.

[0064] In another embodiment, after predicting the main pump flow rate using the saved LSTM model, it further includes: calculating the volumetric efficiency using the predicted main pump flow rate to evaluate whether the predicted flow rate is accurate.

[0065] Further, under the condition of uniform right rotation, the volumetric efficiency = actual flow rate / theoretical flow rate; where the actual flow rate = the LSTM predicted flow rate, and the theoretical flow rate = the rated displacement of the pump × 0.001 (unit conversion) × the measured engine speed.

[0066] Further, the method for evaluating whether the predicted flow rate is accurate is as follows: determine whether the volumetric efficiency of the main pump is within the normal range of 0.95 - 0.98; if so, it proves that the predicted flow rate is accurate.

[0067] Figure 3 The following shows the flow of the early fault diagnosis method for the main pump of a hydraulic shovel based on the LSTM flow prediction model implemented according to the actually obtained multivariate time series data. The specific implementation steps include:

[0068] 1) Collect multivariate time series data at the edge.

[0069] 2) Process the input data, interpolate the low-frequency data to make its frequency consistent with the high-frequency frequency, and align the processed data in time.

[0070] 3) Split the input data into training and testing sets according to the time series, and scale the data using MinMaxScaler; use the input values of the past 30 time points (all features except the target column) to predict the 31st target value, and perform predictions sequentially by sliding the sliding window downward.

[0071] 4) When training the model, perform some hyperparameter micro-adjustments to find the basic model, check the best parameters of the model, and save the best model.

[0072] 5) Use the saved LSTM model to predict the flow rate of the main pump, and inverse-scale the data predicted by the model to obtain the true predicted flow rate data.

[0073] 6) Calculate the volumetric efficiency of the main pump using the flow rate of the main pump predicted by the LSTM model, and use the calculated volumetric efficiency to evaluate whether the prediction process is accurate; determine whether the volumetric efficiency is within the normal range of 0.95 - 0.98, and evaluate whether the predicted flow rate is accurate through the volumetric ratio.

[0074] Please refer to Figure 4 , which shows the structural schematic diagram of the early fault diagnosis device for the main pump of a hydraulic shovel based on the LSTM flow prediction model provided by an exemplary embodiment of the present invention. The device includes a collection module, a data processing module, a test splitting module, a training module, and a prediction module.

[0075] Specifically, the collection module is used to collect data at the edge side and screen the data related to the main pump as input data.

[0076] A data processing module for processing the neatness of the collected input data;

[0077] A test splitting module for training and test splitting of the processed input data;

[0078] A training module for training the LSTM model, obtaining the optimal parameters of the model using girdsearchCV, and saving the optimal model;

[0079] A prediction module for predicting the main pump flow rate using the saved LSTM model.

[0080] Optionally, the device further includes an evaluation module for calculating the volumetric efficiency using the predicted main pump flow rate to evaluate whether the predicted flow rate is accurate.

[0081] It should be noted that the hydraulic shovel main pump early fault diagnosis device based on the LSTM flow prediction model provided in the above embodiment and the embodiment of the hydraulic shovel main pump early fault diagnosis method based on the LSTM flow prediction model belong to the same concept. The implementation process is detailed in the hydraulic shovel main pump early fault diagnosis method based on the LSTM flow prediction model, which will not be elaborated here.

[0082] In a specific embodiment, 'right rotation at a constant speed': Input: right rotation pilot pressure, right rotation proportional valve current, main pump 2 pressure, engine torque, engine speed; Output: main pump 2 flow rate;

[0083] Experimental operation: Read 85,003 experimental data from September 19, 2022, September 21, 2022, and September 23, 2022 for 'right rotation at a constant speed' as the experimental dataset. The comparison chart of the predicted main pump 2 flow rate and the actual flow rate by this method is as Figure 5 shown. The model test is basically fitted through the comparison chart, and it can be observed that this method can effectively predict the main pump 2 flow rate. The volumetric efficiency of the main pump 2 flow rate predicted by the method of the present invention is as Figure 6 shown. Further calculating the average value of the volumetric efficiency is 0.966, within the volumetric efficiency range of the pump. Therefore, this method can be further used for the calculation of the volumetric efficiency under the condition of right rotation at a constant speed.

[0084] It should be clear that the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0085] In the specification provided herein, a large number of specific details are set forth. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail so as not to obscure the understanding of this description.

[0086] The above are only the preferred embodiments of the present invention, and do not impose any form of limitation on the present invention. Although the present invention has been disclosed as above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art of this patent may, without departing from the scope of the technical solution of the present invention, make some changes or modifications using the technical content prompted above into equivalent embodiments of equivalent changes. However, as long as the content does not depart from the technical solution of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. An early fault diagnosis method for the main pump of a hydraulic shovel based on an LSTM flow prediction model, characterized in that This method is based on the LSTM model for prediction and includes: Collect data at the edge side and screen the data related to the main pump as input data; Process the collected input data for neatness; Split the processed input data into training and testing sets; Train the LSTM model, use gridsearchCV to obtain the optimal parameters of the model, and save the optimal model; Use the saved LSTM model to predict the flow rate of the main pump; The step of collecting data at the edge side and screening the data related to the main pump as input data includes: Divide the working conditions by the pilot pressure values of various actions to obtain data under uniform rotation; Screen the multivariate time series data related to the main pump; the multivariate time series data includes: main pump pressure, engine speed, engine torque percentage, right swing pilot pressure, and right swing proportional valve current; Process the collected input data for neatness, including: performing interpolation processing on low-frequency data to make its frequency consistent with that of high-frequency data, and aligning the processed data in time; The step of splitting the processed input data into training and testing sets includes: Split the input data into training and testing sets according to the time series; Use MinMaxScaler to scale the data; Predict the 31st target value through the input values of the past 30 time points, and use the sliding window method to slide down sequentially for prediction; After using the saved LSTM model to predict the flow rate of the main pump, it further includes: calculating the volumetric efficiency using the predicted main pump flow rate to evaluate whether the predicted flow rate is accurate; Under the condition of uniform right swing, the volumetric efficiency = actual flow rate / theoretical flow rate; where the actual flow rate = LSTM predicted flow rate, and the theoretical flow rate = rated displacement of the pump × 0.001 × measured engine speed.

2. The early fault diagnosis method for the main pump of a hydraulic shovel based on the LSTM flow prediction model according to claim 1, wherein When training the LSTM model, if the LSTM model makes more hyperparameter adjustments, add more layers; if the dataset is very large, increase the number of epochs and units in the LSTM model.

3. The early fault diagnosis method for the main pump of a hydraulic shovel based on the LSTM flow prediction model according to claim 2, characterized in that, When using the saved LSTM model to predict the flow rate of the main pump, it is necessary to inverse-scale the data predicted by the model to obtain the true predicted flow rate data.

4. The early fault diagnosis method for the main pump of a hydraulic shovel based on the LSTM flow prediction model according to claim 1, wherein The method for evaluating whether the predicted flow rate is accurate is: judging whether the volumetric efficiency of the main pump is within the normal range of 0.95 - 0.98; If it is, it proves that the predicted flow rate is accurate.

5. A hydraulic shovel main pump early fault diagnosis device based on an LSTM flow prediction model, characterized in that, This device is used to execute the hydraulic shovel main pump early fault diagnosis method based on the LSTM flow prediction model according to any one of claims 1 to 4. This device includes: A collection module, which is used to collect data at the edge side and screen the data related to the main pump as input data; A data processing module, which is used to process the collected input data for neatness; A test splitting module, which is used to split the processed input data into training and testing sets; A training module, which is used to train the LSTM model, use gridsearchCV to obtain the optimal parameters of the model, and save the optimal model; A prediction module, which is used to use the saved LSTM model to predict the flow rate of the main pump.

Images