Model training method, part life prediction method, electronic equipment and engineering machinery

Abstract

The invention provides a model training method, a part life prediction method, electronic equipment and engineering machinery. The training method of the machine learning model for predicting the service life of the part comprises the steps that multi-modal sample data are collected, the multi-modal sample data comprise current operation data of the engineering machine, environment data of the engineering machine and historical operation data of the engineering machine, and the part is located in the engineering machine; extracting sample features associated with the component from the multi-modal sample data; processing the sample features by using a machine learning model to obtain a life prediction result of the part; determining a loss function according to the life prediction result of the part, the life labeling information of the part and a weight matrix in multiple hidden layers of a machine learning model; and training the machine learning model according to the loss function.

Description

Technical Field

[0001] The present disclosure relates to the field of detection, and in particular to a model training method, a component life prediction method, an electronic device and an engineering machinery. Background Art

[0002] With the continuous development of engine technology, predicting the life of key components, such as the remaining life of the piston, is of great significance to improving the reliability and safety of the engine. Existing engine component life prediction methods mainly rely on a single type of data (such as vibration signals, temperature data, etc.) and traditional machine learning models (such as decision trees, MLP (Multilayer Perceptron)) for life prediction. Summary of the invention

[0003] The inventors have noticed that the engine component life prediction methods in the prior art mainly have the following defects.

[0004] 1. Single data dimension: A single data source cannot fully reflect the working status of the engine piston, resulting in large deviations in the prediction results.

[0005] 2. Limited model capabilities: Traditional machine learning models have limited capabilities when processing high-dimensional complex data and find it difficult to capture complex nonlinear relationships.

[0006] 3. Lack of dynamic adaptability: In actual applications, the engine working environment and load conditions vary greatly. The existing models lack dynamic adaptability and are unable to cope with such changes.

[0007] Accordingly, the present disclosure provides a model training method, which enables the trained model to effectively improve the accuracy and real-time performance of component life prediction.

[0008] In a first aspect of the present disclosure, a method for training a machine learning model for predicting component life is provided, comprising: collecting multimodal sample data, wherein the multimodal sample data comprises current operating data of an engineering machinery, environmental data of the engineering machinery, and historical operating data of the engineering machinery, wherein the component is located in the engineering machinery; extracting sample features associated with the component from the multimodal sample data; processing the sample features using the machine learning model to obtain a life prediction result of the component; determining a loss function based on the life prediction result of the component, the life annotation information of the component, and a weight matrix in a multi-layer hidden layer of the machine learning model; and training the machine learning model based on the loss function.

[0009] In some embodiments, determining the loss function includes: determining a first sub-loss function based on the life prediction result of the component and the life marking information of the component; determining a second sub-loss function based on the weight matrix in the multi-layer hidden layer; and determining the loss function based on the weighted sum of the first sub-loss function and the second sub-loss function.

[0010] In some embodiments, determining the second sub-loss function according to the weight matrix in the multi-layer hidden layer includes: determining the second sub-loss function according to the sum of the squares of the L2 norms of the weight matrices of each hidden layer in the multi-layer hidden layer.

[0011] In some embodiments, the training of the machine learning model according to the loss function includes: calculating the model parameters in the current tth iteration according to the loss function Gradient ; Determine the gradient The first-order momentum estimate of , and the gradient The second-order momentum estimate of ; According to the first-order momentum estimate and the first-order momentum decay rate in the t-th iteration, determine the gradient The first-order momentum bias correction value of ; According to the second-order momentum estimate and the second-order momentum decay rate in the t-th iteration, determine the gradient The second-order momentum bias correction value of According to the model parameters , the first-order momentum deviation correction value and the second-order momentum bias correction value , determine the model parameters in the t+1th iteration In the t+1th iteration, the model parameters are used , train the machine learning model.

[0012] In some embodiments, the step of determining the model parameters in the t+1th iteration is Including: calculating the second-order momentum deviation correction value The square root of and the sum of the predetermined parameters to obtain a first intermediate value; Calculate the first-order momentum deviation correction value The ratio of the model parameter to the first intermediate value is calculated to obtain a second intermediate value; and the weighted sum of the second intermediate value to obtain a third intermediate value; calculating the product of the third intermediate value and a predetermined learning rate to obtain a fourth intermediate value; calculating the model parameter The difference between the fourth intermediate value and the fourth intermediate value is used to obtain the model parameters in the t+1th iteration. .

[0013] In some embodiments, the determining of the gradient The first-order momentum bias correction value of The method comprises: calculating the difference between 1 and the first-order momentum decay rate in the t-th iteration to obtain a first difference; calculating the first-order momentum estimate The ratio of the first difference to obtain the first-order momentum deviation correction value .

[0014] In some embodiments, the determining of the gradient The first-order momentum estimate of Includes: Calculate the first-order momentum estimate of the gradient of the model parameters in the t-1th iteration and the gradient The weighted sum of .

[0015] In some embodiments, the first-order momentum estimate The weight of is the first-order momentum decay rate in the t-th iteration, and the first-order momentum estimate The weights and the gradient The sum of the weights is 1.

[0016] In some embodiments, the determining of the gradient The second-order momentum bias correction value of The method comprises: calculating the difference between 1 and the second-order momentum decay rate in the t-th iteration to obtain a second difference; calculating the second-order momentum estimate The ratio of the second difference to obtain the second-order momentum deviation correction value .

[0017] In some embodiments, the determining of the gradient The second-order momentum estimate of Includes: Calculate the second-order momentum estimate of the gradient of the model parameters in the t-1th iteration and the gradient The weighted sum of the squared values ​​of to obtain the second-order momentum estimate .

[0018] In some embodiments, the second-order momentum estimate The weight of is the second-order momentum decay rate in the t-th iteration, and the second-order momentum estimate The weights and the gradient The sum of the squared weights of is 1.

[0019] In some embodiments, in the process of processing the sample features using the machine learning model, a first output value and a second output value of the lth hidden layer are calculated, wherein the first output value is determined by the weight matrix of the lth hidden layer, the output of the l-1th hidden layer and the bias term of the lth hidden layer, and the second output value is determined by the first output value and the predetermined slope parameter, wherein , is the total number of hidden layers; and selects the larger value of the first output value and the second output value as the output of the lth hidden layer.

[0020] In some embodiments, the calculating of the first output value of the lth hidden layer includes: calculating the product of the weight matrix of the lth hidden layer and the output of the l-1th hidden layer to obtain a linear transformation value; calculating the sum of the linear transformation value and the bias term of the lth hidden layer to obtain the first output value.

[0021] In some embodiments, calculating the second output value of the lth hidden layer includes: calculating the product of the first output value and the predetermined slope parameter to obtain the second output value.

[0022] In some embodiments, extracting sample features associated with the component from the multimodal sample data includes: extracting associated feature information associated with the component using a random forest algorithm; and extracting the sample features from the associated feature information using a LASSO algorithm.

[0023] In a second aspect of the present disclosure, a component life prediction method is provided, comprising: collecting multimodal data, wherein the multimodal data comprises current operating data of an engineering machinery, environmental data of the engineering machinery, and historical operating data of the engineering machinery, wherein the component is located in the engineering machinery; extracting feature information associated with the component from the multimodal data; and processing the feature information using a machine learning model to obtain a life prediction result of the component, wherein the machine learning model is trained using the training method described in any of the above embodiments.

[0024] In a third aspect of the present disclosure, an electronic device is provided, comprising: a memory; and a processor, coupled to the memory, wherein the processor is configured to execute a method as described in any of the above embodiments based on instructions stored in the memory.

[0025] In a fourth aspect of the present disclosure, there is provided an engineering machine comprising the electronic device as described in any of the above embodiments.

[0026] In a fifth aspect of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method described in any of the above embodiments is implemented.

[0027] In a sixth aspect of the present disclosure, a computer program product is provided, comprising computer instructions, wherein when the computer instructions are executed by a processor, the method described in any of the above embodiments is implemented.

[0028] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0030] Figure 1 A flowchart of a method for training a machine learning model for predicting component life according to an embodiment of the present disclosure;

[0031] Figure 2 A schematic diagram of a process flow of a component life prediction method according to an embodiment of the present disclosure;

[0032] Figure 3 A schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure;

[0033] Figure 4 The figure is a schematic diagram of the structure of an engineering machinery according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present disclosure and its application or use. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0035] Unless specifically stated otherwise, the relative arrangement of components and steps, the numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present disclosure.

[0036] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0037] Technologies, methods, and apparatus known to ordinary technicians in the relevant field may not be discussed in detail, but where appropriate, such technologies, methods, and apparatus should be considered part of the authorization specification.

[0038] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0039] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0040] Figure 1 The flowchart of the training method of the machine learning model for predicting the life of a component according to an embodiment of the present disclosure is shown in FIG. In some embodiments, the following training method is performed by an electronic device and includes steps 11-15.

[0041] In step 11 , multimodal sample data is collected, wherein the multimodal sample data includes current operation data of the engineering machine, environmental data of the engineering machine, and historical operation data of the engineering machine, wherein the component is located in the engineering machine.

[0042] For example, the components include engine pistons.

[0043] In some embodiments, the current operating data of the construction machinery includes accelerator pedal, torque, rotation speed, intake pressure, temperature, etc.

[0044] For example, necessary sensors such as cylinder pressure sensors are added to the engine, and combined with relevant sensors of the entire vehicle to collect and store data in real time.

[0045] In some embodiments, the environmental data of the construction machinery includes temperature, humidity, etc.

[0046] For example, the environmental data acquisition method may include at least one of the following.

[0047] 1) Obtain environmental data such as temperature, humidity, and air pressure of the engine environment through environmental sensors.

[0048] 2) The engine’s operating condition and geographic location data are determined through the machine’s operating condition recognition camera and GPS (Global Positioning System).

[0049] It should be noted that these data serve as additional inputs to help the model understand the impact of the environment on the engine operating state.

[0050] In some embodiments, past operating parameters and fault data can be extracted from the historical operating records of the engine. These data provide a reference for the long-term operating status of the engine and help identify and predict future faults.

[0051] In some embodiments, the acquired data may be subjected to data cleaning, data denoising, data normalization, and the like to facilitate subsequent model training.

[0052] 1) Data cleaning

[0053] The purpose of data cleaning is to remove or correct errors, missing values ​​or outliers in the data, thereby improving data quality.

[0054] 1.1) Process missing values. For example, use formula (1) or formula (2) to fill in missing values.

[0055] (1)

[0056] That is, for missing values , mean filling is achieved through formula (1).

[0057] (2)

[0058] That is, for time series data, interpolation can be performed using formula (2) to supplement missing values.

[0059] 1.2) Dealing with outliers: Detect and remove outliers through statistical methods such as z-score or IQR (interquartile range).

[0060] (3)

[0061] In the z-score method shown in formula (3), if or ,but are considered outliers.

[0062] (4)

[0063] In the IQR method shown in formula (4), if the data point is not in the range If it is within, it is considered as an outlier.

[0064] 2) Data denoising

[0065] The purpose of denoising is to reduce the noise in the data and enhance the clarity of the signal. Common methods include wavelet transform and moving average filtering.

[0066] 2.1) Wavelet transform: Wavelet transform can effectively separate noise from signals by decomposing them into components of different frequencies.

[0067] (5)

[0068] In formula (5), is the mother wavelet function, is the wavelet coefficient. Noise can be removed by selecting an appropriate threshold λ, as shown in formula (6).

[0069] (6)

[0070] 2.2) Moving average filtering: By calculating the average value of the data in the sliding window, data fluctuations are smoothed and high-frequency noise is reduced.

[0071] (7)

[0072] In formula (7), is the smoothed signal at time t, and n is the window size.

[0073] 3) Data normalization

[0074] The purpose of normalization is to scale data of different features to the same scale to prevent the magnitude difference of feature values ​​from affecting model training.

[0075] Min-Max normalization is shown in formula (8).

[0076] (8)

[0077] In formula (8), It is the normalized data, usually in the range of [0,1].

[0078] Z-score normalization is shown in formula (9).

[0079] (9)

[0080] In formula (9), is the mean, is the standard deviation. Z-score standardization transforms the data into a distribution with a mean of 0 and a standard deviation of 1.

[0081] In step 12, sample features associated with the components are extracted from the multimodal sample data.

[0082] In some embodiments, a random forest algorithm is used to extract associated feature information associated with a component, and a LASSO (Least Absolute Shrinkage and Selection Operator) algorithm is used to extract sample features from the associated feature information.

[0083] The following is a brief description of the random forest algorithm and the LASSO algorithm.

[0084] 1) Feature importance screening of random forests:

[0085] When the characteristics When it is used to split a node, the reduction in the Gini index can be calculated using formula (10).

[0086] (10)

[0087] In formula (10), D is the data set of the current node, and D1 and D2 are the two subsets obtained after splitting.

[0088] Similarly, the information gain can be calculated by using formula (11).

[0089] (11)

[0090] In formula (11), H(D) is the entropy of the data set of the current node, and H(D1) and H(D2) are the entropies of the two subsets after splitting.

[0091] Whenever the feature is used to split a node, and its importance contribution is accumulated. For a tree t, feature Total importance contribution As shown in formula (12) or formula (13).

[0092] (12)

[0093] (13)

[0094] In formula (12) and formula (13), Represents the tree t using features The set of all nodes to be split.

[0095] feature The overall importance of the contribution is the average of its importance contribution in all trees, as shown in formula (14).

[0096] (14)

[0097] In formula (14), T is the number of decision trees in the random forest, It is a feature Importance contribution in the t-th tree.

[0098] Through the above processing, the features with higher importance are preliminarily screened out for the next step of calculation.

[0099] 2) Feature selection and model optimization of LASSO:

[0100] When optimizing the regression model, features are selected by adding L1 regularization. The objective function of the LASSO optimization problem is shown in formula (15).

[0101] (15)

[0102] In formula (15), N is the number of samples, p is the number of features, is the jth eigenvalue of the ith sample, is the target value, is the regression coefficient of the jth feature, is the regularization parameter. LASSO will adjust value, so that the unimportant features correspond to is zero, thus performing feature selection.

[0103] In step 13, the sample features are processed using a machine learning model to obtain a component life prediction result.

[0104] In some embodiments, in the process of processing sample features using a machine learning model, a first output value and a second output value of the lth hidden layer are calculated, wherein the first output value is determined by the weight matrix of the lth hidden layer, the output of the l-1th hidden layer and the bias term of the lth hidden layer, and the second output value is determined by the first output value and a predetermined slope parameter, wherein , Next, the larger value between the first output value and the second output value is selected as the output of the lth hidden layer.

[0105] In some embodiments, the product of the weight matrix of the lth hidden layer and the output of the l-1th hidden layer is calculated to obtain a linear transformation value. Next, the sum of the linear transformation value and the bias term of the lth hidden layer is calculated to obtain a first output value.

[0106] In some embodiments, the product of the first output value and a predetermined slope parameter is calculated to obtain the second output value.

[0107] For example, the output of the lth hidden layer As shown in formula (16).

[0108] (16)

[0109] In formula (16), is the weight matrix of the lth hidden layer, is the output of the l-1th hidden layer, is the bias term of the lth hidden layer, is the slope parameter of Leak ReLu, which is used to handle the situation where the input is less than 0.

[0110] It should be noted that after all hidden layers are calculated, the calculation of the unimodal output layer is shown in formula (17).

[0111] (17)

[0112] In formula (17), is the final output of the i-th model, is the weight matrix of the output layer, is the output of the Lth hidden layer, is the bias term of the output layer.

[0113] In step 14, a loss function is determined based on the life prediction result of the component, the life annotation information of the component, and the weight matrix in the multi-layer hidden layer of the machine learning model.

[0114] In some embodiments, a first sub-loss function is determined based on the life prediction result of the component and the life annotation information of the component. A second sub-loss function is determined based on the weight matrix in the multi-layer hidden layer. A loss function is determined based on the weighted sum of the first sub-loss function and the second sub-loss function.

[0115] In some embodiments, the second sub-loss function is determined according to the sum of squares of the L2 norms of the weight matrices of each hidden layer in the multiple hidden layers.

[0116] For example, the loss function As shown in formula (18).

[0117] (18)

[0118] In formula (18), N is the total number of single-mode models, is the output result of the i-th unimodal model, is the annotation information corresponding to the output result of the ith unimodal model, M is the total number of hidden layers in the machine learning model, is the weight matrix of the jth hidden layer, is the regularization coefficient.

[0119] It should be noted here that, is the regularization term.

[0120] By introducing the regularization term, the loss function not only considers the prediction error, but also controls the complexity of the model to avoid overfitting.

[0121] In step 15, the machine learning model is trained according to the loss function.

[0122] It should be noted here that during the training process, back propagation updates the model parameters by calculating the gradient of the loss function for each model parameter, so that the loss function is gradually reduced.

[0123] In some embodiments, the step of training the machine learning model according to the loss function includes the following.

[0124] 1) Calculate the model parameters in the current tth iteration according to the loss function Gradient .

[0125] In some embodiments, the gradient As shown in formula (19).

[0126] (19)

[0127] In formula (19), is the loss function right gradient.

[0128] 2) Determine the gradient The first-order momentum estimate of , and the gradient The second-order momentum estimate of .

[0129] In some embodiments, the first-order momentum estimate of the gradient of the model parameters in the t-1th iteration is calculated and gradient to obtain the first-order momentum estimate .

[0130] In some embodiments, the first-order momentum estimate The weight is the first-order momentum decay rate in the tth iteration , and the first-order momentum estimate The weight of and gradient The sum of the weights is 1.

[0131] For example, the gradient The first-order momentum estimate of As shown in formula (20).

[0132] (20)

[0133] In formula (20), the first-order momentum decay rate is Can be set to 0.9.

[0134] In some embodiments, the second-order momentum estimate of the gradient of the model parameters in the t-1th iteration is calculated and gradient The weighted sum of the squared values ​​of to obtain the second-order momentum estimate .

[0135] In some embodiments, the second-order momentum estimate The weight is the second-order momentum decay rate in the tth iteration , and the second-order momentum estimate The weight of and gradient The sum of the squared weights of is 1.

[0136] For example, the gradient The second-order momentum estimate of As shown in formula (21).

[0137] (twenty one)

[0138] In formula (21), the second-order momentum decay rate is Can be set to 0.999.

[0139] 3) According to the first-order momentum estimate and the first-order momentum decay rate in the tth iteration, determine the gradient The first-order momentum bias correction value of .

[0140] In some embodiments, the difference between 1 and the first-order momentum decay rate in the t-th iteration is calculated to obtain a first difference. Next, the first-order momentum estimate is calculated The ratio of the first difference to obtain the first-order momentum deviation correction value .

[0141] For example, the first-order momentum bias correction value As shown in formula (22)

[0142] (twenty two)

[0143] 4) Based on the second-order momentum estimate and the second-order momentum decay rate in the tth iteration, determining the gradient The second-order momentum bias correction value of .

[0144] In some embodiments, the difference between 1 and the second-order momentum decay rate in the t-th iteration is calculated to obtain a second difference. Next, the second-order momentum estimate is calculated The ratio of the second difference to obtain the second-order momentum deviation correction value .

[0145] For example, the second-order momentum bias correction value As shown in formula (23).

[0146] (twenty three)

[0147] 5) According to the model parameters , first-order momentum deviation correction value and the second-order momentum bias correction value , determine the model parameters in the t+1th iteration .

[0148] In some embodiments, the model parameters in the t+1th iteration are determined The steps include the following.

[0149] 5.1) Calculate the second-order momentum bias correction value The square root of and the sum of a predetermined parameter to obtain a first intermediate value.

[0150] 5.2) Calculate the first-order momentum bias correction value The ratio of the first intermediate value to the second intermediate value is obtained.

[0151] 5.3) Calculate model parameters The weighted sum of and the second intermediate value is used to obtain the third intermediate value.

[0152] 5.4) Calculate the product of the third intermediate value and the predetermined learning rate to obtain a fourth intermediate value.

[0153] 5.5) Calculate model parameters The difference between the fourth intermediate value and the fourth intermediate value is used to obtain the model parameters in the t+1th iteration. .

[0154] For example, the model parameters As shown in formula (24).

[0155] (twenty four)

[0156] In formula (24), is the weight decay coefficient, which controls the decay rate of each parameter. is a small amount (usually set to ). is the learning rate.

[0157] It should be noted here that by separating weight decay and learning rate, the complexity of the model can be more effectively controlled.

[0158] Through the above-mentioned adaptive optimization method, the parameter update can be effectively realized.

[0159] 6) In the t+1th iteration, use the model parameters , to train the machine learning model.

[0160] In the model training method provided in the above embodiment of the present disclosure, multi-source data is integrated to automatically extract key features, and an optimized deep neural network model is used to perform life prediction, thereby improving prediction accuracy and stability.

[0161] Figure 2 The flowchart of the component life prediction method according to one embodiment of the present disclosure is shown in FIG. 1. In some embodiments, the following training method is executed by an electronic device, including steps 21-23.

[0162] In steps 21 , 22 , the component is located in the construction machine.

[0163] For example, the components include engine pistons.

[0164] In some embodiments, the current operation data of the engineering machine includes the accelerator pedal, torque, rotation speed, intake pressure, temperature, etc. The environmental data of the engineering machine includes temperature, humidity, etc.

[0165] In some embodiments, past operating parameters and fault data can be extracted from the historical operating records of the engine. These data provide a reference for the long-term operating status of the engine and help identify and predict future faults.

[0166] In step 22 , feature information associated with the component is extracted from the multimodal data.

[0167] In some embodiments, the random forest algorithm is used to extract the associated feature information associated with the component, and the LASSO algorithm is used to extract the sample features from the associated feature information.

[0168] In step 23, the feature information is processed using a machine learning model to obtain a life prediction result of the component, wherein the machine learning model uses Figure 1 The training is obtained by training using the training method involved in any one of the embodiments.

[0169] It should be noted here that the above adaptive optimization process can be repeated based on the comparison between the predicted value and the true value to improve the matching degree of the parameters.

[0170] For example, for the multimodal model, the final prediction value is shown in formula (25).

[0171] (25)

[0172] set up is the prediction result of the ith sub-model, For its weight.

[0173] During the dynamic adjustment process, adjust according to the current error or other evaluation indicators , as shown in formula (26).

[0174]

[0175] in, It can be calculated based on the model's prediction error or time-weighted error, so that the most recent prediction results have a greater impact on the final output.

[0176] In addition, the prediction results can be visualized and alarmed.

[0177] 1) Visualization of prediction results:

[0178] - Use visualization tools (such as line charts and heat maps) to display life prediction results, allowing users to intuitively understand the health status of engine pistons.

[0179] - Visualization results can include predicted remaining life curves, changing trends of key features, etc.

[0180] 2) Perform alarm processing based on the prediction results:

[0181] - Set a life threshold. When the remaining life of the piston predicted by the model is lower than the set threshold, the system automatically triggers an alarm to remind the user to perform necessary maintenance operations.

[0182] - Alarm information can be sent to relevant personnel via SMS, email or console prompts, with maintenance suggestions attached.

[0183] Figure 3 The figure is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure.

[0184] like Figure 3 As shown, the electronic device 30 may be in the form of a general-purpose computing device and includes a memory 31, a processor 32, and a bus 33 for connecting different system components.

[0185] The memory 31 may include, for example, a system memory, a non-volatile storage medium, etc. The system memory may store, for example, an operating system, an application program, a boot loader, and other programs. The system memory may include a volatile storage medium, such as a random access memory (RAM) and / or a cache memory. The non-volatile storage medium may store, for example, instructions of a corresponding embodiment of at least one model training method and a component life prediction method being executed. The non-volatile storage medium includes, but is not limited to, a disk storage, an optical storage, a flash memory, and the like.

[0186] The processor 32 may be implemented by a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistors, and other discrete hardware components. Accordingly, each module such as an acquisition module, a calculation module, and an adjustment module may be implemented by a central processing unit (CPU) running instructions in a memory that execute corresponding steps, or may be implemented by a dedicated circuit that executes corresponding steps.

[0187] For example, the processor 32 is configured to execute instructions stored in the memory to implement the following Figure 1 or Figure 2 The method of any one of the embodiments.

[0188] The bus 33 may use any of a variety of bus structures, including, but not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MCA) bus, and a Peripheral Component Interconnect (PCI) bus.

[0189] The interfaces 34, 35, 36 of the electronic device 30, the memory 31 and the processor 32 can be connected via a bus 33. The input / output interface 34 provides a connection interface for input / output devices such as a display, a mouse, and a keyboard. The network interface 35 provides a connection interface for various networked devices. The storage interface 36 provides a connection interface for external storage devices such as a floppy disk, a USB flash drive, and an SD card.

[0190] Here, various aspects of the present disclosure are described with reference to flowcharts and / or block diagrams of methods, devices, and computer program products according to embodiments of the present disclosure. It should be understood that each frame of the flowchart and / or block diagram and the combination of frames can be implemented by computer-readable program instructions.

[0191] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable device to produce a machine, so that the processor executes the instructions to produce means for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.

[0192] These computer-readable program instructions may also be stored in a computer-readable memory, which cause the computer to work in a specific manner to produce an article of manufacture, including instructions for implementing the functions specified in one or more blocks in the flowchart and / or block diagram.

[0193] The present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects.

[0194] The present disclosure also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the following is implemented: Figure 1 or Figure 2 The method of any one of the embodiments.

[0195] The present disclosure also provides a computer program product, including computer instructions, wherein when the computer instructions are executed by a processor, the following is achieved: Figure 1 or Figure 2 The method of any one of the embodiments.

[0196] Figure 4 The figure is a schematic diagram of the structure of an engineering machinery according to an embodiment of the present disclosure.

[0197] like Figure 4 As shown, the engineering machine 40 includes an electronic device 41. The electronic device 41 is Figure 3 An electronic device according to any one of the embodiments.

[0198] By implementing the above-mentioned embodiments of the present disclosure, the following beneficial effects can be obtained.

[0199] 1) It can more accurately estimate and observe the usage status of the engine piston.

[0200] 2) The method for predicting piston life can be replicated to other components.

[0201] 3) It can effectively optimize the design of components based on the operating conditions of the entire machine and the operating conditions of the engine.

[0202] 4) Improve the quality and efficiency of after-sales service.

[0203] In some embodiments, the functional unit described above may be implemented as a general-purpose processor, a programmable logic controller (PLC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any appropriate combination thereof, for performing the functions described in the present disclosure.

[0204] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0205] The description of the present disclosure is given for the purpose of illustration and description, and is not intended to be exhaustive or to limit the present disclosure to the disclosed form. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present disclosure, and to enable those of ordinary skill in the art to understand the present disclosure and thereby design various embodiments with various modifications suitable for specific uses.

Claims

1. A method for training a machine learning model for predicting component life, comprising: collecting multimodal sample data, wherein the multimodal sample data includes current operation data of an engineering machine, environmental data of the engineering machine, and historical operation data of the engineering machine, wherein the component is located in the engineering machine; extracting sample features associated with the component from the multimodal sample data; Processing the sample features using the machine learning model to obtain a life prediction result of the component; Determine a loss function according to the life prediction result of the component, the life annotation information of the component, and the weight matrix in the multi-layer hidden layer of the machine learning model; The machine learning model is trained according to the loss function.

2. The training method according to claim 1, wherein: Determining the loss function includes: Determining a first sub-loss function according to the life prediction result of the component and the life annotation information of the component; Determining a second sub-loss function according to the weight matrix in the multi-layer hidden layer; The loss function is determined according to a weighted sum of the first sub-loss function and the second sub-loss function.

3. The training method according to claim 2, wherein: Determining the second sub-loss function according to the weight matrix in the multi-layer hidden layer includes: The second sub-loss function is determined according to the square sum of the L2 norms of the weight matrices of each hidden layer in the multiple hidden layers.

4. The training method according to claim 1, wherein: The training of the machine learning model according to the loss function comprises: Calculate the model parameters in the current t-th iteration according to the loss function Gradient ; Determine the gradient The first-order momentum estimate of , and the gradient The second-order momentum estimate of ; According to the first-order momentum estimate and the first-order momentum decay rate in the t-th iteration, determine the gradient The first-order momentum bias correction value of ; According to the second-order momentum estimate and the second-order momentum decay rate in the t-th iteration, determine the gradient The second-order momentum bias correction value of ; According to the model parameters , the first-order momentum deviation correction value and the second-order momentum bias correction value , determine the model parameters in the t+1th iteration ; In the t+1th iteration, the model parameters are used , train the machine learning model.

5. The training method according to claim 4, wherein: Determining the model parameters in the t+1th iteration include: Calculate the second-order momentum bias correction value The square root of and the sum of a predetermined parameter to obtain a first intermediate value; Calculate the first-order momentum bias correction value to the first intermediate value to obtain a second intermediate value; Calculate the model parameters and the second intermediate value to obtain a third intermediate value; Calculating a product of the third intermediate value and a predetermined learning rate to obtain a fourth intermediate value; Calculate the model parameters The difference between the fourth intermediate value and the fourth intermediate value is used to obtain the model parameters in the t+1th iteration. .

6. The training method according to claim 4, wherein: Determining the gradient The first-order momentum bias correction value of include: Calculate the difference between 1 and the first-order momentum decay rate in the t-th iteration to obtain a first difference value; Calculate the first-order momentum estimate The ratio of the first difference to obtain the first-order momentum deviation correction value .

7. The training method according to claim 4, wherein: Determining the gradient The first-order momentum estimate of include: Calculate the first-order momentum estimate of the gradient of the model parameters in the t-1th iteration and the gradient The weighted sum of .

8. The training method according to claim 7, wherein: The first-order momentum estimate The weight of is the first-order momentum decay rate in the t-th iteration, and the first-order momentum estimate The weights and the gradient The sum of the weights is 1.

9. The training method according to claim 4, wherein: Determining the gradient The second-order momentum bias correction value of include: Calculate the difference between 1 and the second-order momentum decay rate in the t-th iteration to obtain a second difference; Calculate the second-order momentum estimate The ratio of the second difference to obtain the second-order momentum deviation correction value .

10. The training method according to claim 4, wherein: Determining the gradient The second-order momentum estimate of include: Calculate the second-order momentum estimate of the gradient of the model parameters in the t-1th iteration and the gradient The weighted sum of the squared values ​​of to obtain the second-order momentum estimate .

11. The training method according to claim 10, wherein: The second-order momentum estimate The weight of is the second-order momentum decay rate in the t-th iteration, and the second-order momentum estimate The weights and the gradient The sum of the squared weights of is 1.

12. The training method according to claim 1, further comprising: In the process of processing the sample features using the machine learning model, a first output value and a second output value of the lth hidden layer are calculated, wherein the first output value is determined by the weight matrix of the lth hidden layer, the output of the l-1th hidden layer and the bias term of the lth hidden layer, and the second output value is determined by the first output value and the predetermined slope parameter, wherein , is the total number of hidden layers; Select the larger value between the first output value and the second output value as the output of the lth hidden layer.

13. The training method according to claim 12, wherein: The calculating of the first output value of the lth hidden layer comprises: Calculate the product of the weight matrix of the lth hidden layer and the output of the l-1th hidden layer to obtain a linear transformation value; The sum of the linear transformation value and the bias term of the l-th hidden layer is calculated to obtain the first output value.

14. The training method according to claim 13, wherein: The calculating of the second output value of the lth hidden layer comprises: The product of the first output value and the predetermined slope parameter is calculated to obtain the second output value.

15. The training method according to any one of claims 1 to 14, wherein: Extracting sample features associated with the component from the multimodal sample data includes: Using a random forest algorithm, extracting associated feature information associated with the component; The sample features are extracted from the associated feature information using the LASSO algorithm.

16. A component life prediction method, comprising: collecting multimodal data, wherein the multimodal data includes current operating data of a construction machine, environmental data of the construction machine, and historical operating data of the construction machine, wherein the component is located in the construction machine; extracting feature information associated with the component from the multimodal data; The characteristic information is processed using a machine learning model to obtain a life prediction result of the component, wherein the machine learning model is trained using the training method described in any one of claims 1-15.

17. An electronic device comprising: Memory; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 1 to 16 based on instructions stored in the memory.

18. An engineering machine comprising the electronic device according to claim 17.

19. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to any one of claims 1 to 16 is implemented.

20. A computer program product, comprising computer instructions, wherein the computer instructions implement the method according to any one of claims 1 to 16 when executed by a processor.

Images