Engine residual life prediction method and model based on neural network

Abstract

The invention relates to the technical field of excavator engine health management, in particular to an engine residual life prediction method and model based on a neural network, and the method comprises the following steps: S101, an Internet of Things terminal collects data; s102, the gateway receives data uploaded by the terminal; s103, the cloud platform processes the data into feature data; and S104, substituting the feature data into a residual life prediction model based on a neural network, and returning the predicted life duration of the equipment by the model. The residual life prediction model is characterized by mainly comprising an input layer, a convolutional neural network (CNN) layer, a long short-term memory (LSTM) network layer and a full connection layer, according to the invention, various sensor signals generated during operation of the engine equipment are acquired by adopting the Internet of Vehicles technology, and the residual service life of the engine is predicted through the neural network model, so that the equipment maintenance strategy can be optimized, the downtime caused by equipment failure is reduced, and the continuity and stability of production are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of engine health management of excavators, and in particular to a method and model for predicting the remaining life of an engine based on a neural network. Background Art

[0002] Excavators are a commonly used type of construction machinery, widely used in earthwork, mining, urban construction, water conservancy, and other fields. For the health management of excavator engines, the main means of equipment operation and maintenance is regular maintenance and repair. There are also some technologies that predict the remaining service life of equipment parts (bearings) through vibration parameters or crack shape and size.

[0003] Existing technologies either use high-frequency vibration signals (millisecond level) or use computer vision to identify cracks in bearings for prediction, which does not conform to the actual working scenarios of construction machinery (vibration signals require continuous high-frequency vibration data, which can only be collected on-site or transmitted at high frequency, and the data collection cost is high). In actual application, there is no essential difference from regular maintenance, and advance prediction cannot be achieved. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method and model for predicting the remaining service life of an engine based on a neural network. The parameters such as the intake manifold temperature, urea temperature, engine water temperature, engine intake temperature, engine exhaust temperature and the like in the entire operating life cycle of the equipment and their life cycle are used as a data set to train the model and predict the remaining service life of the engine.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A method for predicting the remaining life of an engine based on a neural network, the steps are as follows:

[0007] S101, IoT terminal collects data: IoT terminal collects various sensor signals of the excavator engine and uploads them to IoT gateway;

[0008] S102, the gateway receives the data uploaded by the terminal: the IoT gateway receives the data uploaded by the terminal, associates the information uploaded by the IoT terminal with the excavator equipment, and stores it in the IoT data warehouse;

[0009] S103, the cloud platform processes the data into feature data: the IoT cloud platform extracts relevant sensor parameter signal values ​​from the data warehouse and performs normalization processing to produce feature data;

[0010] S104. The feature data is brought into a remaining life prediction model based on a neural network, and the model returns the predicted life span of the equipment.

[0011] Furthermore, the standardization and normalization processing in step S103 includes Z score standardization processing and min-max normalization processing.

[0012] Furthermore, the Z score normalization process subtracts the mean of each operating condition parameter value and divides it by the standard deviation of the parameter value. The formula is as follows:

[0013] .

[0014] Furthermore, the min-max normalization process maps the data to the interval (0,1), subtracts the minimum value of each parameter from the minimum value of the parameter, and divides it by the difference between the maximum and minimum values ​​of the parameter. The normalization formula is as follows:

[0015] .

[0016] Further, a neural network-based engine remaining life prediction model is characterized in that it is mainly composed of an input layer, a convolutional neural network layer (CNN), a long short-term memory network layer (LSTM) and a fully connected layer;

[0017] The convolutional neural network layer includes a convolution layer and a pooling layer; the convolution layer uses a sliding filter window to perform a convolution operation on the input device feature data to extract the features of the data; the pooling layer downsamples the data of the convolution layer to reduce the dimension of the data and the number of parameters to prevent overfitting;

[0018] The long short-term memory network layer controls the flow of information in the network by introducing a gating mechanism, thereby effectively processing long sequence data;

[0019] The fully connected layer highly abstracts the convolutional layer, pooling layer, and long short-term memory network layer into a one-dimensional feature vector. Each node is connected to all the nodes in the previous layer, and the input of each node is the output of all the nodes in the previous layer. After weighted summation and adding the bias term, a nonlinear transformation is performed through the activation function to introduce nonlinear factors and enhance the expressive power of the neural network.

[0020] Further, the process of remaining life prediction model training is as follows:

[0021] S201, equipment life data set: first, it is necessary to collect sensor signal data marked with remaining service life labels in the entire life cycle of multiple engines that have reached the end of their service life to form an equipment life data set;

[0022] S202, data preprocessing: standardizing and normalizing the data of the equipment life data set;

[0023] S203, data set grouping: dividing the standardized equipment life data set into a training set and a test set in a ratio of 8:2;

[0024] S204, constructing a neural network model: constructing a neural network model mainly composed of an input layer, a convolutional neural network layer, a long short-term memory network layer and a fully connected layer;

[0025] S205. Train the neural network: Use random values ​​to initialize the weights and biases in the network, input the training set into the model, obtain the output value, compare it with the true label, calculate the loss value under the current parameters according to the loss function, and use the gradient descent algorithm to update the weights and biases in the model to minimize the loss function; repeat this process until the loss value is less than the set threshold.

[0026] S206, test data set: bring the test set into the model and calculate the root mean square error (RMSE) between the predicted value and the true value to evaluate the model;

[0027] S207: Life prediction model: After the above steps, a remaining life prediction model based on a neural network is obtained.

[0028] Furthermore, the RMSE formula in step S206 is:

[0029]

[0030] Where N is the number of test set data, T is the true value, and P is the model prediction value.

[0031] Beneficial effects: The present invention uses vehicle networking technology to collect various sensor signals generated during the operation of engine equipment, and predicts the remaining service life of the engine through a neural network model, which can optimize equipment maintenance strategies and avoid unnecessary costs and time waste caused by premature or late maintenance; improve production efficiency and economic benefits, reduce downtime caused by equipment failure, and ensure production continuity and stability; the present invention has good scalability, and the relevant models can be adjusted and optimized to adapt to engines of different models and brands. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is the overall architecture of the engine remaining useful life prediction system in the embodiment of the present invention.

[0033] Figure 2 This is the overall process of training the neural network model for predicting the remaining useful life of an engine in an embodiment of the present invention.

[0034] Figure 3 It is a long short-term memory network (LSTM) structure in an embodiment of the present invention.

[0035] Figure 4It is a convolutional neural network-long short-term memory network structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations on the technical solution of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0037] The term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0038] like Figure 1 As shown, the present invention describes a method for predicting the remaining service life of an engine based on a neural network, which collects operating numerical features such as engine intake manifold temperature, urea temperature, engine water temperature, engine intake temperature, engine exhaust temperature, and corresponding label remaining service life, and establishes a convolutional neural network (CNN)-long short-term memory neural network (LSTM) model. After the real-time collected data is sent to the edge-cloud collaborative gateway through the Internet of Things terminal, the Internet of Things cloud platform obtains a predicted value through the model input feature value, thereby realizing the prediction of the remaining service life of the engine.

[0039] S101: IoT terminal data collection

[0040] Various types of sensors are installed in different parts of the engine equipment, which are transmitted to the IoT terminal through CAN bus communication. The IoT terminal will collect various sensor signals of the excavator's engine (intake manifold temperature, urea temperature, engine water temperature, engine intake temperature, engine exhaust temperature, etc.), package and encode them uniformly, and upload them to the IoT gateway.

[0041] S102: Gateway receives data uploaded by terminal

[0042] The IoT gateway receives the data uploaded by the terminal, and associates the information uploaded by the IoT terminal with the excavator equipment through the binding relationship of the IoT cloud platform. The IoT cloud platform obtains the operating conditions of the engine equipment and stores it in the IoT data warehouse.

[0043] S103: The cloud platform processes the data into feature data

[0044] The IoT cloud platform periodically extracts the relevant sensor parameter values ​​(intake manifold temperature, urea temperature, engine water temperature, engine intake temperature, engine exhaust temperature, etc.) in the previous day's operating data from the data warehouse, and performs Zscore standardization on each parameter value, that is, subtracting the mean of each operating parameter value and dividing it by the standard deviation of the parameter value, as shown in formula (1):

[0045] (1)

[0046] The equipment operating data after Z score standardization can greatly reduce the differences caused by changes in different working scenarios.

[0047] Then, min-max normalization is used to map the data to the (0,1) interval. The normalization method is shown in formula (2):

[0048] (2)

[0049] That is, for each standardized parameter, each parameter value is subtracted from the minimum value of the parameter and divided by the difference between the maximum and minimum values ​​of the parameter.

[0050] S104: Substitute feature data into the neural network model

[0051] The normalized characteristic data of S103 is brought into the training of the remaining life prediction model based on the neural network (the process is as follows Figure 2 The model returns the predicted life span of the equipment.

[0052] The key technical solutions of the present invention include equipment remaining life data set collection, data preprocessing, data set grouping, neural network model construction, neural network model training, and data set testing. The model obtained through training can realize the prediction of the remaining service life of the engine.

[0053] The overall training process of the neural network-based engine remaining life prediction model used by the IoT cloud platform is as follows: Figure 2 shown.

[0054] S201: Equipment life data set

[0055] The neural network model requires a large amount of data sets for training. First, it is necessary to collect the sensor signals and corresponding usage time of the engines of multiple excavators that have reached the end of their service life. The end of the engine's life is subtracted from the usage time in the life cycle to obtain the sensor signal data marked with the remaining service life label (the equipment regularly uploads various parameters and the corresponding cumulative engine working time) to form the equipment life data set.

[0056] S202: Data preprocessing

[0057] Similar to S103, the training model also needs to perform the same standardization and normalization processing on the data of the equipment life data set.

[0058] S203: Dataset Grouping

[0059] The standardized equipment life data set is divided into a training set and a test set in a ratio of 8:2. The training set is used to train the data model, and the test set is used to test the accuracy of the model.

[0060] S204: Building a neural network model

[0061] The model structure adopted by the present invention is as follows Figure 4 As shown in the figure, it is mainly composed of an input layer, a convolutional neural network layer (CNN, including a convolutional layer and a pooling layer), a long short-term memory network layer (LSTM) and a fully connected layer.

[0062] The convolution layer uses a sliding filter window to perform convolution operations on the input device feature data to extract the features of the data. The pooling layer downsamples the data of the convolution layer to reduce the dimension of the data and the number of parameters to prevent overfitting.

[0063] The long short-term memory network layer controls the flow of information in the network by introducing a gating mechanism (including forget gate, input gate, and output gate), so that LSTM can retain information when necessary and discard irrelevant information, thereby effectively processing long sequence data. Its structure is as follows Figure 3 shown.

[0064] The fully connected layer highly abstracts the convolutional layer, pooling layer, and LSTM layer into a one-dimensional feature vector, and each node is connected to all nodes in the previous layer. The input of each node is the output of all nodes in the previous layer. After weighted summation and adding the bias term, a nonlinear transformation is performed through the activation function to introduce nonlinear factors and enhance the expression ability of the neural network.

[0065] S205: Training Neural Networks

[0066] Use random values ​​to initialize the weights and biases in the network, input the training set into the model, obtain the output value, compare it with the true label, calculate the loss value under the current parameters according to the loss function, and use the gradient descent algorithm to update the weights and biases in the model to minimize the loss function.

[0067] This process is repeated until the loss value is less than the set threshold.

[0068] S206: Test Dataset

[0069] The test set is introduced into the model, and the root mean square error (RMSE) between the predicted value and the true value is calculated to evaluate the model. The definition of RMSE is shown in formula (3):

[0070] (3)

[0071] Where N is the number of test set data, T is the true value, and P is the model prediction value.

[0072] S207: Lifespan prediction model

[0073] The obtained neural network model is the remaining life prediction model based on neural network in the embodiment of the present invention.

[0074] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of an engine based on a neural network, characterized in that: S101, IoT terminal collects data: IoT terminal collects various sensor signals of the excavator engine and uploads them to IoT gateway; S102, the gateway receives the data uploaded by the terminal: the IoT gateway receives the data uploaded by the terminal, associates the information uploaded by the IoT terminal with the excavator equipment, and stores it in the IoT data warehouse; S103, the cloud platform processes the data into feature data: the IoT cloud platform extracts relevant sensor parameter signal values ​​from the data warehouse and performs normalization processing to produce feature data; S104. The feature data is brought into a remaining life prediction model based on a neural network, and the model returns the predicted life span of the equipment.

2. The method for predicting the remaining life of an engine based on a neural network according to claim 1, characterized in that: The standardization and normalization processing in step S103 includes Z score standardization processing and min-max normalization processing.

3. The method for predicting the remaining life of an engine based on a neural network according to claim 2, characterized in that: The Z score standardization process subtracts the mean of each operating condition parameter value and divides it by the standard deviation of the parameter value. The formula is as follows: 。 4. The method for predicting the remaining life of an engine based on a neural network according to claim 2, characterized in that: The min-max normalization process maps the data to the interval (0,1), subtracts the minimum value of each parameter from the minimum value of the parameter, and divides it by the difference between the maximum and minimum values ​​of the parameter. The normalization formula is as follows: 。 5. A neural network-based engine remaining life prediction model, characterized in that: It is mainly composed of input layer, convolutional neural network layer (CNN), long short-term memory network layer (LSTM) and fully connected layer; The convolutional neural network layer includes a convolution layer and a pooling layer; the convolution layer uses a sliding filter window to perform a convolution operation on the input device feature data to extract the features of the data; the pooling layer downsamples the data of the convolution layer to reduce the dimension of the data and the number of parameters to prevent overfitting; The long short-term memory network layer controls the flow of information in the network by introducing a gating mechanism, thereby effectively processing long sequence data; The fully connected layer highly abstracts the convolutional layer, pooling layer, and long short-term memory network layer into a one-dimensional feature vector. Each node is connected to all the nodes in the previous layer, and the input of each node is the output of all the nodes in the previous layer. After weighted summation and adding the bias term, a nonlinear transformation is performed through the activation function to introduce nonlinear factors and enhance the expressive power of the neural network.

6. The neural network-based engine remaining life prediction model according to claim 5, characterized in that: The training process: S201, equipment life data set: first, it is necessary to collect sensor signal data marked with remaining service life labels in the entire life cycle of multiple engines that have reached the end of their service life to form an equipment life data set; S202, data preprocessing: standardizing and normalizing the data of the equipment life data set; S203, data set grouping: dividing the standardized equipment life data set into a training set and a test set in a ratio of 8:2; S204, constructing a neural network model: constructing a neural network model mainly composed of an input layer, a convolutional neural network layer, a long short-term memory network layer and a fully connected layer; S205. Train the neural network: Use random values ​​to initialize the weights and biases in the network, input the training set into the model, obtain the output value, compare it with the true label, calculate the loss value under the current parameters according to the loss function, and use the gradient descent algorithm to update the weights and biases in the model to minimize the loss function; repeat this process until the loss value is less than the set threshold.

7. S206, test data set: bring the test set into the model and calculate the root mean square error (RMSE) between the predicted value and the true value to evaluate the model; S207: Life prediction model: after the above steps, a remaining life prediction model based on a neural network is obtained; The neural network-based engine remaining life prediction model according to claim 6 is characterized in that: The RMSE formula in step S206 is: ， Where N is the number of test set data, T is the true value, and P is the model prediction value.

Images