Milling cutter wear prediction method and system for nonmetal carrier roller shaft machining

Abstract

The invention provides a milling cutter abrasion prediction method and system for nonmetal carrier roller shaft machining, and relates to the technical field of nonmetal carrier roller shaft machining cutter abrasion prediction. Preprocessing the basic data set; based on a time domain feature extraction algorithm combining CEEMDAN decomposition and statistical calculation, deep feature information is extracted; sorting the N feature variables from large to small based on the importance score of each feature variable by using a random forest algorithm, and generating a model training data set; and training the neural network model according to the model training data set, and outputting the abrasion loss of the to-be-predicted milling cutter at the target moment through the trained neural network model according to the input operation parameters corresponding to the target moment of the to-be-predicted milling cutter. According to the method, the wear condition of the milling cutter can be efficiently and accurately predicted, decision support is provided for maintenance and replacement of the milling cutter in the production process of the non-metal roller carrier shaft, and therefore the production efficiency and the product quality are improved.

Description

Technical Field

[0001] This invention relates to the field of tool wear prediction technology for non-metallic idler roller shaft machining, and in particular to a method and system for predicting milling cutter wear in non-metallic idler roller shaft machining. Background Technology

[0002] In the field of non-metallic idler roller shaft processing, the milling process places stringent requirements on the performance of milling cutters. Milling cutters are prone to wear under high-load, high-precision machining environments, and the degree of wear directly affects the machining accuracy and production efficiency of the idler roller shaft. Therefore, timely and accurate monitoring and prediction of milling cutter wear has become a key aspect of ensuring stable production of non-metallic idler roller shafts. With the continuous development of processing technology, enterprises are increasingly demanding intelligent and automated production processes. Traditional methods of milling cutter wear detection, which rely on manual intervention, are no longer sufficient to meet the needs of modern production. A more efficient technical solution is urgently needed to address the problem of milling cutter wear prediction.

[0003] Currently, the industry mainly attempts to solve the problem of milling cutter wear detection through two technical approaches: one is a combination of manual inspection and dimensional measurement, where technicians periodically stop the machine to perform visual inspections of the milling cutters and use measuring tools to measure key dimensions to determine the wear condition; the other is a single-signal-based monitoring technology, which uses sensors to collect vibration or temperature signals during the milling process and uses simple signal analysis methods to determine whether the milling cutter is in an abnormal wear state. These two technologies provide support for milling cutter wear monitoring to some extent and have been applied in some simple machining scenarios.

[0004] However, the aforementioned existing technologies have significant technical shortcomings, namely, they are difficult to achieve real-time and accurate prediction of milling cutter wear under the complex working conditions of non-metallic idler shaft machining. Manual inspection and dimensional measurement require machine downtime, which not only interrupts the production process and reduces production efficiency, but also relies on human experience and judgment, which is highly subjective and cannot capture subtle changes in milling cutter wear. Monitoring technologies based on single signals can only obtain partial data information during the machining process, and cannot fully mine the deep features of the data. Moreover, under complex working conditions, the signal is easily affected by noise, resulting in insufficient effectiveness of the extracted features. Ultimately, the milling cutter wear prediction results are of low accuracy and poor stability, making it difficult to meet the requirements of high-precision milling cutter wear prediction for non-metallic idler shaft machining. Summary of the Invention

[0005] This invention provides a method and system for predicting milling cutter wear in the machining of non-metallic idler roller shafts. It can efficiently and accurately predict the wear of milling cutters, providing decision support for the maintenance and replacement of milling cutters in the production process of non-metallic idler roller shafts, thereby improving production efficiency and product quality.

[0006] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: Firstly, a method for predicting milling cutter wear in the machining of non-metallic idler shafts is provided. The method includes: acquiring a basic dataset generated during the machining process of the milling cutter on the non-metallic idler shaft, the basic dataset including the operating parameters and wear amount of the milling cutter at each of multiple time points; preprocessing the basic dataset to obtain a preprocessed dataset, the preprocessing including outlier removal and normalization; and extracting deep feature information from the preprocessed dataset based on a time-domain feature extraction algorithm combining CEEMDAN decomposition and statistical calculation, the deep feature information including N feature variables corresponding to each of multiple time points, each feature variable representing a different dimension of time-domain features; and utilizing... The importance score of each feature variable is determined using a random forest algorithm, and the N feature variables are sorted from largest to smallest based on their importance scores to generate a model training dataset. This dataset includes the top Q feature variables and their wear values ​​at each of multiple time points. The importance score is positively correlated with the degree of influence on the wear of the milling cutter, Q is less than N, and both Q and N are positive integers. The neural network model is trained using this dataset to obtain a trained neural network model. Based on the input operating parameters corresponding to the target time of the milling cutter to be predicted, the trained neural network model outputs the wear value of the milling cutter to be predicted at the target time.

[0007] The beneficial effects of this invention are as follows: The method provided in this embodiment constructs a basic dataset by using multi-time-lapse parameters and wear amount, ensuring that the data covers the temporal process of milling cutter operation; the outlier removal and normalization processing in the preprocessing stage solves the problems of high noise in the original data and the impact of dimensional differences on model training, laying a high-quality data foundation for subsequent feature extraction; the feature extraction method combining CEEMDAN decomposition and statistical calculation can mine deep temporal features in non-stationary operating parameters, making up for the shortcomings of traditional single feature extraction in capturing subtle wear signals of milling cutters; the random forest algorithm filters highly important features, which can eliminate redundant information, reduce model complexity, and improve model training efficiency. Wear prediction is achieved based on the trained neural network model, which can output accurate results by utilizing the complex correlation between deep features and wear amount, ultimately realizing online and accurate prediction of milling cutter wear, solving the problems of low efficiency and poor prediction accuracy of traditional manual stop detection, and ensuring the continuity of non-metallic idler shaft processing and product accuracy.

[0008] In one possible implementation of the first aspect, the operating parameters include spindle speed, feed rate, depth of cut, width of cut, cutting force signals and vibration signals of the milling machine in the X, Y, and Z axes.

[0009] The method provided in this invention overcomes the limitations of traditional wear prediction, which relies on only a single type of parameter, by limiting operating parameters including spindle speed, feed rate, depth of cut, width of cut, X / Y / Z axis cutting force signals, and vibration signals. The combination of these two types of parameters comprehensively reflects the entire process of active control and passive feedback in milling cutter operation, avoiding the omission of key wear-related information due to missing parameters. This ensures the completeness of the subsequent basic dataset and the effectiveness of feature extraction, thereby improving the reliability of model prediction and ensuring that wear trend misjudgments are not caused by incomplete parameter coverage.

[0010] In one possible implementation of the first aspect, acquiring the basic dataset generated by the milling cutter during the machining of the non-metallic idler roller shaft includes: The spindle speed, feed rate, depth of cut, and width of cut are obtained at each moment in multiple time periods through the computer numerical control system of the milling machine. The cutting force and vibration signals of the milling machine in the X, Y, and Z axes are acquired by sensors. The wear amount at each time point in multiple time periods was obtained using a super depth-of-field 3D microscope; The basic dataset is generated based on the spindle speed, feed rate, depth of cut, width of cut, cutting force signals in the X, Y, and Z axes, vibration signals, and wear amount of the milling machine at each time point in multiple moments.

[0011] The methods provided in this invention acquire parameters through different devices or equipment. First, acquiring process parameters through a CNC system avoids errors from manual measurement, ensuring the accuracy and real-time nature of the parameters. Second, the cutting force and vibration signals acquired by sensors capture the dynamic changes during the milling cutter's cutting process, providing direct evidence of wear status and solving the problem of traditional reliance on static parameters making it difficult to capture real-time wear. Finally, the wear amount acquired through a super-depth-of-field 3D microscope is a high-precision measured value, which is more accurate than traditional caliper measurements and can provide accurate label values ​​for the model. The combination of these three professional parameter acquisition methods ensures that the basic dataset is accurate, real-time, and highly precise, avoiding subsequent model training biases caused by data quality issues from the source, and laying a data foundation for the accuracy of wear prediction.

[0012] In one possible implementation of the first aspect, the preprocessing of the base dataset to obtain a preprocessed dataset includes: Determine the standard deviation and mean of each running parameter at each time step in the base dataset; Not located The operating parameters within the specified range were identified as outliers; Each outlier is replaced by the running parameters corresponding to the M time points before each outlier, resulting in a dataset after outlier removal, where M is a positive integer; Normalize each running parameter in the dataset after removing outliers to obtain a preprocessed dataset; The calculation formula for the normalization process is as follows: ; in, For the dataset after removing outliers, For the preprocessed dataset, and These are the minimum and maximum values ​​in the dataset after outliers have been removed.

[0013] The method provided in this invention identifies outlier data and replaces it with the average of the parameters from the previous M time steps. This avoids the disruption of time-series data caused by directly deleting outliers, while simultaneously using historical reasonable data to correct deviations, ensuring the temporal continuity and rationality of the data. This solves the problem of traditional outlier handling destroying data integrity. Furthermore, in the normalization process, a formula eliminates the dimensional differences between different parameters, ensuring fair weighting of each parameter during model training. This prevents any single parameter from dominating the model learning process due to excessive magnitude, improving the stability and convergence speed of model training. The preprocessed dataset has low noise and a uniform format, providing high-quality input for subsequent feature extraction and model training, reducing prediction errors caused by data interference.

[0014] In one possible implementation of the first aspect, the temporal feature extraction algorithm based on CEEMDAN decomposition and statistical computation extracts deep feature information from the preprocessed dataset, including: Based on a time-domain feature extraction algorithm combining CEEMDAN decomposition and statistical calculation, each operating parameter corresponding to each time step in the preprocessed dataset is decomposed to obtain t intrinsic mode functions and residuals corresponding to each operating parameter. Based on statistical calculation methods, multiple statistical parameters corresponding to each intrinsic mode function and residual are determined. These statistical parameters include sum, minimum, maximum, mean, standard deviation, variance, root mean square error, skewness, kurtosis, peak factor, and impulse factor. The deep feature information corresponding to each operating parameter is obtained by using the intrinsic mode function and multiple statistical parameters of the residual corresponding to each operating parameter at each time.

[0015] The method provided in this invention firstly decomposes non-stationary operating parameters into t intrinsic mode functions and residuals using the CEEMDAN algorithm, separating different frequency components and avoiding feature ambiguity caused by the mixing of signals of different frequencies. Secondly, it characterizes the attributes of each decomposed component from three dimensions—centralized features, discrete features, and distribution shape features—using multiple statistical parameters, which can capture subtle features of milling cutter wear. This combination of decomposition and multi-dimensional statistics can extract deep information strongly correlated with milling cutter wear from operating parameters, avoiding the shortcomings of traditional single statistics in reflecting wear trends, and providing rich and effective deep feature variables for subsequent feature selection.

[0016] In one possible implementation of the first aspect, determining the importance score of each feature variable using the random forest algorithm includes: For characteristic variables The random forest algorithm is used to process each decision tree. Select the corresponding data outside the bag. Calculate the error outside the bag ,in, ; Random pair of feature variables Incorporate noise interference and calculate out-of-bag data error. ; Based on n bag outside errors Errors in data outside the bag Determine feature variables Importance score; feature Importance score The calculation formula is: .

[0017] The method provided in this invention first calculates the error using out-of-bag data through a random forest algorithm, eliminating the need for additional validation set partitioning, thus fully utilizing existing data and avoiding data waste. Then, it quantifies feature importance by comparing the out-of-bag errors before and after adding noise. That is, adding noise to important features significantly increases the model's prediction error for wear amount; this method objectively reflects the actual impact of features on the prediction results, avoiding the subjective bias of traditional feature selection based on experience. Third, it filters features based on importance scores, retaining the top Q high-impact features. This eliminates redundant features while ensuring the model focuses on key information, preventing redundant features from interfering with the model's learning process. Ultimately, this improves model training efficiency and prediction accuracy, ensuring the model can output reliable wear prediction results based on core features.

[0018] In one possible implementation of the first aspect, the neural network model is a deep neural network model, and the hidden layer of the neural network model is l layers; The output of the i-th neuron in the k-th hidden layer of the neural network model The calculation formula is: ; in, The number of neurons in the (k-1)th hidden layer is defined as The output of the j-th neuron in the (k-1)-th hidden layer is , It is the connection weight between the j-th neuron in the (k-1)-th hidden layer and the i-th neuron in the k-th hidden layer, and This represents the bias of the i-th neuron in the k-th hidden layer. This represents the tanh activation function, and its specific calculation formula is as follows: .

[0019] The method provided in this invention addresses the problem of traditional shallow neural networks struggling to learn complex feature relationships by limiting the neural network to a deep neural network (DNN) and designing multiple hidden layers and a tanh activation function. Firstly, the single hidden layer can gradually uncover the complex mapping relationship between deep features and milling cutter wear through multiple rounds of nonlinear transformations, avoiding the shortcomings of shallow networks in fitting the data. Secondly, the tanh activation function has a wider output range and can effectively handle negative changes in features. Compared to the traditional sigmoid function, it is more suitable for dynamic feature processing of milling cutter operating parameters, ensuring that the model does not lose key negative feature information during the learning process. The combination of the DNN's structural design and the activation function selection enhances the model's ability to fit milling cutter wear under complex working conditions, ensuring accurate prediction results even when parameters change dynamically.

[0020] In one possible implementation of the first aspect, the output expression y of the neural network model is: ; Where b is the bias of the output layer neuron, and the variable is... This represents the output of the i-th neuron in the l-th hidden layer, while the variable... This represents the weight between the i-th neuron in the l-th hidden layer and the neuron in the output layer.

[0021] The method provided in this invention ensures the practicality and rationality of the prediction results by limiting the calculation logic of the output layer of the DNN model: the output layer calculates the predicted value by multiplying the output of the l-th hidden layer by the weights and the bias, which comprehensively outputs the predicted value based on the importance of each hidden layer feature, avoiding the dominance of a single feature in the output result; at the same time, this formula ensures that the output wear prediction value falls within the reasonable range of reality, avoiding the abnormal prediction values ​​caused by non-standard output layer design in traditional models. This improves the operability of the prediction method, ensuring that the model output is not only accurate but also effectively guides the processing and production of non-metallic idler rollers.

[0022] Secondly, the present invention provides a milling cutter wear prediction system for machining non-metallic idler roller shafts. The system includes: an acquisition module for acquiring a basic dataset generated during the machining of non-metallic idler roller shafts, the basic dataset including the milling cutter's operating parameters and wear amount at each of multiple time points; a preprocessing module for preprocessing the basic dataset to obtain a preprocessed dataset, the preprocessing including outlier removal and normalization; and a feature extraction module for extracting deep feature information from the preprocessed dataset based on a time-domain feature extraction algorithm combining CEEMDAN decomposition and statistical calculations, the deep feature information including N feature variables corresponding to each of multiple time points, each feature variable representing a different dimension of time. The system comprises: a domain feature module; a training data generation module, used to determine the importance score of each feature variable using a random forest algorithm, and to sort the N feature variables from largest to smallest based on the importance score of each feature variable to generate a model training dataset, wherein the model training dataset includes the top Q feature variables and wear amount corresponding to each time step at multiple time steps, the importance score being positively correlated with the degree of influence on the wear amount of the milling cutter, Q being less than N, and Q and N being positive integers; and a prediction module, used to train the neural network model based on the model training dataset to obtain the trained neural network model; and to output the wear amount of the milling cutter to be predicted at the target time based on the input operating parameters corresponding to the target time of the milling cutter through the trained neural network model.

[0023] Thirdly, an electronic device is provided, the electronic device including a memory and one or more processors; the memory is coupled to the processors; wherein the memory stores computer program code, the computer program code including computer instructions, which, when executed by the processor, cause the electronic device to perform the method as described in any implementation of the first aspect.

[0024] Fourthly, a computer-readable storage medium is provided, including computer instructions that, when executed on an electronic device, cause the electronic device to perform a method as described in any implementation of the first aspect.

[0025] Fifthly, a computer program product is provided that, when run on a computer, causes the computer to perform the method in any implementation of the first aspect.

[0026] Understandably, the beneficial effects achieved by the system of the second aspect, the electronic device of the third aspect, the computer-readable storage medium of the fourth aspect, and the computer program product of the fifth aspect provided above can be referred to with reference to the beneficial effects of the first aspect and any of its possible design embodiments, which will not be repeated here. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention; Figure 2 A flowchart of a milling cutter wear prediction method for non-metallic idler roller shaft machining provided in an embodiment of the present invention; Figure 3 A flowchart of another milling cutter wear prediction method for non-metallic roller shaft machining provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a prediction system provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be described below with reference to the accompanying drawings. In the description of the present invention, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can represent A or B. The "or" in the present invention is merely a description of the relationship between the related objects, indicating that three relationships can exist. For example, A or B can represent: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Furthermore, in the description of the present invention, unless otherwise stated, "multiple" refers to two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.

[0029] Furthermore, to facilitate a clear description of the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0030] In this embodiment of the invention, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this embodiment of the invention should not be construed as superior or more advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner for ease of understanding.

[0031] In the field of non-metallic idler roller shaft processing, the milling process places stringent requirements on the performance of milling cutters. Milling cutters are prone to wear under high-load, high-precision machining environments, and the degree of wear directly affects the machining accuracy and production efficiency of the idler roller shaft. Therefore, timely and accurate monitoring and prediction of milling cutter wear has become a key aspect of ensuring stable production of non-metallic idler roller shafts. With the continuous development of processing technology, enterprises are increasingly demanding intelligent and automated production processes. Traditional methods of milling cutter wear detection, which rely on manual intervention, are no longer sufficient to meet the needs of modern production. A more efficient technical solution is urgently needed to address the problem of milling cutter wear prediction.

[0032] Currently, the industry mainly attempts to solve the problem of milling cutter wear detection through two technical approaches: one is a combination of manual inspection and dimensional measurement, where technicians periodically stop the machine to perform visual inspections of the milling cutters and use measuring tools to measure key dimensions to determine the wear condition; the other is a single-signal-based monitoring technology, which uses sensors to collect vibration or temperature signals during the milling process and uses simple signal analysis methods to determine whether the milling cutter is in an abnormal wear state. These two technologies provide support for milling cutter wear monitoring to some extent and have been applied in some simple machining scenarios.

[0033] However, the aforementioned existing technologies have significant technical shortcomings, namely, they are difficult to achieve real-time and accurate prediction of milling cutter wear under the complex working conditions of non-metallic idler shaft machining. Manual inspection and dimensional measurement require machine downtime, which not only interrupts the production process and reduces production efficiency, but also relies on human experience and judgment, which is highly subjective and cannot capture subtle changes in milling cutter wear. Monitoring technologies based on single signals can only obtain partial data information during the machining process, and cannot fully mine the deep features of the data. Moreover, under complex working conditions, the signal is easily affected by noise, resulting in insufficient effectiveness of the extracted features. Ultimately, the milling cutter wear prediction results are of low accuracy and poor stability, making it difficult to meet the requirements of high-precision milling cutter wear prediction for non-metallic idler shaft machining.

[0034] Therefore, embodiments of the present invention provide a method and system for predicting milling cutter wear in the machining of non-metallic idler shafts. The method includes: acquiring a basic dataset generated by the milling cutter during the machining of non-metallic idler shafts, the basic dataset including the operating parameters and wear amount of the milling cutter at each of multiple time points; preprocessing the basic dataset to obtain a preprocessed dataset, the preprocessing including outlier removal and normalization; and extracting deep feature information from the preprocessed dataset based on a time-domain feature extraction algorithm combining CEEMDAN decomposition and statistical calculation, the deep feature information including N feature variables at each of multiple time points, each feature variable representing a different dimension of the time domain. Features; The importance score of each feature variable is determined using a random forest algorithm, and the N feature variables are sorted from largest to smallest based on their importance scores to generate a model training dataset. This dataset includes the top Q feature variables and wear amount at each of multiple time points. The importance score is positively correlated with the degree of influence on the wear amount of the milling cutter, Q is less than N, and both Q and N are positive integers. The neural network model is trained using the training dataset to obtain a trained neural network model. Based on the input operating parameters corresponding to the target time of the milling cutter to be predicted, the trained neural network model outputs the wear amount of the milling cutter to be predicted at the target time.

[0035] The method provided in this invention constructs a basic dataset using multi-time-lapse parameters and wear amounts, ensuring data coverage of the milling cutter's temporal process. Outlier removal and normalization in the preprocessing stage address the issues of high noise and dimensional differences in the original data affecting model training, laying a high-quality data foundation for subsequent feature extraction. The feature extraction method combining CEEMDAN decomposition and statistical calculations can uncover deep temporal features in non-stationary operating parameters, overcoming the limitations of traditional single feature extraction in capturing subtle milling cutter wear signals. The random forest algorithm filters highly important features, eliminating redundant information, reducing model complexity, and improving model training efficiency. Wear prediction is achieved based on the trained neural network model, utilizing the complex correlation between deep features and wear amounts to output accurate results. Ultimately, online and accurate prediction of milling cutter wear is achieved, solving the problems of low efficiency and poor prediction accuracy associated with traditional manual stop-and-detection, ensuring the continuity of non-metallic roller shaft processing and product precision.

[0036] In some embodiments, the milling cutter wear prediction method for non-metallic idler shaft machining provided by the present invention can be executed by a milling cutter wear prediction system 100 for non-metallic idler shaft machining (hereinafter referred to as prediction system 100).

[0037] As an example, the prediction system 100 can be any electronic device 200 with data processing capabilities, such as a general-purpose computer, personal computer, laptop computer, switch, or tablet computer. The specific implementation of the prediction system 100 is not limited here.

[0038] For example, the non-metallic idler roller shaft can be made of Q235 material. Q235 is a type of carbon structural steel widely used in construction, machinery manufacturing, and other fields. Here, Q represents yield strength, and 235 indicates that its yield strength value is not less than 235 MPa, which is a core indicator for measuring the steel's resistance to deformation.

[0039] Figure 1 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention is shown. The electronic device 200 includes a processor 210, a memory 220, and a communication interface 230.

[0040] Processor 210 may include one or more processing cores. Processor 210 connects to various parts within electronic device 200 using various interfaces and lines, and performs various functions and processes data of electronic device 200 by running or executing instructions, programs, code sets, or instruction sets stored in memory 220, and by calling data stored in memory 220. Optionally, processor 210 may be implemented using at least one of the following hardware forms: Central Processing Unit (CPU), Graphics Processing Unit (GPU), Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA).

[0041] The memory 220 may include random access memory (RAI) or read-only memory (ROI). Optionally, the memory 220 may include non-transitory computer-readable storage ledger. The memory 220 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 220 may include a stored program area. The stored program area may store instructions for implementing an operating system, instructions for implementing at least one function, instructions for implementing the various method embodiments described above, etc.

[0042] Communication interface 230 is used to communicate with other devices, equipment or communication networks, such as data storage devices, image processing devices or Ethernet, wireless access network (RAN), wireless local area network (WLAN), etc.

[0043] In terms of physical implementation, the aforementioned devices (such as processor 210, memory 220, and communication interface 230) can each be devices within the same device (such as a laptop computer). Alternatively, at least two of these devices can be located within the same device, i.e., as different devices within the same device, similar to the deployment of devices or components in a distributed system.

[0044] It is understood that the structure illustrated in this embodiment does not constitute a specific limitation on the electronic device 200. In other embodiments of the present invention, the electronic device 200 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0045] The following description, in conjunction with the accompanying drawings, illustrates a milling cutter wear prediction method for machining non-metallic roller shafts provided by an embodiment of the present invention.

[0046] Figure 2 This is a flowchart illustrating a milling cutter wear prediction method for machining non-metallic roller shafts, provided as an embodiment of the present invention. Optionally, this method can be... Figure 1 The prediction system 100 shown is... Figure 2 The illustrated electronic device 200 performs this operation. The method may include the following steps: S1. Obtain the basic dataset generated by the milling cutter during the machining of non-metallic idler roller shafts. The basic dataset includes the operating parameters and wear amount of the milling cutter at each time point in multiple moments.

[0047] In one possible implementation, the operating parameters include spindle speed, feed rate, depth of cut, width of cut, cutting force signals and vibration signals of the milling machine in the X, Y, and Z axes.

[0048] The method provided in this invention overcomes the limitations of traditional wear prediction, which relies on only a single type of parameter, by limiting operating parameters including spindle speed, feed rate, depth of cut, width of cut, X / Y / Z axis cutting force signals, and vibration signals. The combination of these two types of parameters comprehensively reflects the entire process of active control and passive feedback in milling cutter operation, avoiding the omission of key wear-related information due to missing parameters. This ensures the completeness of the subsequent basic dataset and the effectiveness of feature extraction, thereby improving the reliability of model prediction and ensuring that wear trend misjudgments are not caused by incomplete parameter coverage.

[0049] In one possible implementation, S1 includes: acquiring the spindle speed, feed rate, depth of cut, and width of cut at each of multiple moments using the computer numerical control system of the milling machine; acquiring cutting force signals and vibration signals of the milling machine in the X, Y, and Z axes using sensors; acquiring the wear amount at each of multiple moments using a super depth-of-field 3D microscope; and generating the basic dataset based on the spindle speed, feed rate, depth of cut, width of cut, cutting force signals, vibration signals, and wear amount at each of multiple moments.

[0050] Specifically, the basic dataset include: ; Where MS is the spindle speed and FR is the feed rate. C is the depth of cut, CW is the width of cut, and the cutting force signals in the X, Y, and Z axes are respectively... , The vibration signals in the X, Y, and Z axes are , , , This refers to wear and tear. In other words, the underlying dataset... It contains 11 runtime variables, among which This represents the spindle speed at each of the n time points. This represents the amount of wear at each of the n time points.

[0051] The methods provided in this invention acquire parameters through different devices or equipment. First, acquiring process parameters through a CNC system avoids errors from manual measurement, ensuring the accuracy and real-time nature of the parameters. Second, the cutting force and vibration signals acquired by sensors capture the dynamic changes during the milling cutter's cutting process, providing direct evidence of wear status and solving the problem of traditional reliance on static parameters making it difficult to capture real-time wear. Finally, the wear amount acquired through a super-depth-of-field 3D microscope is a high-precision measured value, which is more accurate than traditional caliper measurements and can provide accurate label values ​​for the model. The combination of these three professional parameter acquisition methods ensures that the basic dataset is accurate, real-time, and highly precise, avoiding subsequent model training biases caused by data quality issues from the source, and laying a data foundation for the accuracy of wear prediction.

[0052] S2. Preprocess the basic dataset to obtain a preprocessed dataset. The preprocessing includes removing outliers and normalization.

[0053] In some embodiments, S2 includes: Determine the standard deviation and mean of each running parameter at each time step in the base dataset; [the dataset will then be divided into groups not included in the dataset]. The operating parameters within the specified range are identified as outliers; each outlier is replaced based on the operating parameters corresponding to the M time points preceding it, resulting in a dataset with outliers removed, where M is a positive integer; each operating parameter in the dataset with outliers removed is normalized to obtain a preprocessed dataset; the normalization calculation formula is as follows: ; in, For the dataset after removing outliers, For the preprocessed dataset, and These are the minimum and maximum values ​​in the dataset after outliers have been removed.

[0054] In one example, M=20.

[0055] The method provided in this invention identifies outlier data and replaces it with the average of the parameters from the previous 20 time steps. This avoids the data breakage caused by directly deleting outliers, while simultaneously using historical reasonable data to correct biases, ensuring the continuity and rationality of the data's temporal sequence. This solves the problem of traditional outlier handling destroying data integrity. Furthermore, in the normalization process, a formula eliminates the dimensional differences between different parameters, ensuring fair weighting of each parameter during model training. This prevents any single parameter from dominating the model learning process due to excessive magnitude, improving the stability and convergence speed of model training. The preprocessed dataset has low noise and a uniform format, providing high-quality input for subsequent feature extraction and model training, reducing prediction errors caused by data interference.

[0056] S3. A time-domain feature extraction algorithm based on CEEMDAN decomposition and statistical calculation is used to extract deep feature information from the preprocessed dataset.

[0057] The deep feature information includes N feature variables corresponding to each time point in multiple time points, and each feature variable is used to characterize time-domain features of different dimensions.

[0058] In some embodiments, see Figure 3 The above S3 includes: S31. A time-domain feature extraction algorithm based on CEEMDAN decomposition and statistical calculation is used to decompose each operating parameter corresponding to each time step in the preprocessed dataset to obtain t intrinsic mode functions and residuals corresponding to each operating parameter.

[0059] Specifically, the following explanation uses the decomposition of spindle speed MS as an example. The decomposed representation of spindle speed MS is as follows: ; in, Dataset This represents the first intrinsic mode function of the spindle speed decomposition at the first time step. This represents the first residual of the spindle speed decomposition at the first moment. express The feature vector is decomposed into Each intrinsic mode function.

[0060] Specifically, the prediction system uses the CEEMDAN algorithm to perform frequency-level processing on each operating parameter at each time step. The preprocessed operating parameters (such as spindle speed MS, X-axis cutting force FX, etc.) are essentially non-stationary signals. For example, spindle speed is affected by load changes due to cutter wear, and simultaneously contains high-frequency instantaneous fluctuations (such as cutting impact) and low-frequency stable trends (such as the normal operating baseline). Direct analysis is prone to losing wear-related information due to signal contamination. The CEEMDAN algorithm can decompose this non-stationary signal from high to low frequency, obtaining t intrinsic mode functions (IMFs, representing signal components of different frequencies) and one residual (representing the overall trend of the signal). Taking spindle speed MS as an example, if at time step 1, t = 3 IMFs and 1 residual are obtained, then in the matrix... It is the first high-frequency IMF (corresponding to the instantaneous impulse signal) decomposed from the MS at that moment. This is the residual of MS at that moment (corresponding to the stable operating trend); the MS at the 2nd moment and the nth moment are also decomposed according to the same logic. Finally, each operating parameter at each moment is broken down into multi-dimensional components, laying the foundation for the subsequent accurate extraction of wear features.

[0061] S32. Determine multiple statistical parameters corresponding to each intrinsic mode function and residual based on statistical calculation methods.

[0062] The statistical parameters include sum, minimum, maximum, mean, standard deviation, variance, root mean square deviation, skewness, kurtosis, peak factor, and impulse factor.

[0063] Specifically, the prediction system quantifies the features of each IMF and residual after decomposition using multi-dimensional statistical parameters. After the above decomposition steps, each IMF and residual corresponds to a time series data (e.g., IMF1 of MS at time 1 includes multiple data points before and after that time), but these data still need to be converted into feature values ​​that can be recognized by the model. At this time, the features are quantified from three dimensions using 11 statistical parameters: central tendency features (such as mean and sum) reflect the overall level of the component, discrete features (such as standard deviation and variance) reflect the degree of data fluctuation, and distribution shape features (such as skewness and kurtosis) reflect the asymmetry or steepness of the data distribution. These three dimensions can precisely cover the key correlation information of milling cutter wear. For example, when the milling cutter wear intensifies, a certain IMF component of the X-axis cutting force FX (corresponding to the cutting load fluctuation) will show changes such as an increase in the maximum value, an increase in the standard deviation, and a deviation of the kurtosis from the normal range. These changes can be accurately captured by statistical parameters, avoiding the shortcomings of traditional methods that only use the average value to reflect the details of fluctuations.

[0064] S33. Based on the intrinsic mode function and multiple statistical parameters of the residual corresponding to each operating parameter at each time, obtain the deep feature information corresponding to each operating parameter.

[0065] Specifically, the prediction system integrates all quantized features to form deep feature information for each operating parameter. For each time step, an operating parameter, after the above steps, yields (t+1) components (t IMFs + 1 residual). Each component, after the above steps, yields 11 statistical parameters. Therefore, the number of deep features for that operating parameter at that time step is (t+1) × 11. If there are 10 operating parameters (such as MS, FR, FX, FY, etc.), then the total number of feature variables at that time step is N = 10 × (t+1) × 11. For example, when t = 3 (3 IMFs) and the operating parameter is MS, the deep features of MS at time step 1 include 11 parameters from IMF1, 11 parameters from IMF2, 11 parameters from IMF3, and 11 parameters from the residuals, totaling 44 feature variables. Integrating these feature variables from all operating parameters forms the deep feature information for that time step. The feature information from multiple time steps together constitutes a complete deep feature dataset. These feature variables can comprehensively characterize the relationship between operating parameters and milling cutter wear from different frequencies and dimensions, providing rich and effective inputs for subsequent random forest selection of key features and DNN model accurate prediction.

[0066] The method provided in this invention firstly decomposes non-stationary operating parameters into t intrinsic mode functions and residuals using the CEEMDAN algorithm, separating different frequency components and avoiding feature ambiguity caused by the mixing of signals of different frequencies. Secondly, it characterizes the attributes of each decomposed component from three dimensions—centralized features, discrete features, and distribution shape features—using multiple statistical parameters, which can capture subtle features of milling cutter wear. This combination of decomposition and multi-dimensional statistics can extract deep information strongly correlated with milling cutter wear from operating parameters, avoiding the shortcomings of traditional single statistics in reflecting wear trends, and providing rich and effective deep feature variables for subsequent feature selection.

[0067] S4. Use the random forest algorithm to determine the importance score of each feature variable, and sort the N feature variables from largest to smallest based on the importance score of each feature variable to generate the model training dataset.

[0068] The model training dataset includes the top Q feature variables and wear amount at each time step across multiple time steps. The importance score is positively correlated with the degree of influence on the wear amount of the milling cutter, Q is less than N, and Q and N are positive integers.

[0069] In some embodiments, S4 above includes: For characteristic variables The random forest algorithm is used to process each decision tree. Select the corresponding data outside the bag. Calculate the error outside the bag ,in, Randomly select the characteristic variable Incorporate noise interference and calculate out-of-bag data error. Based on n out-of-bag errors Errors in data outside the bag Determine feature variables Importance score; feature Importance score The calculation formula is: .

[0070] Specifically, after the feature extraction in S3, N deep feature variables are generated at each moment (such as IMF components of different operating parameters, statistical parameters corresponding to residuals). However, there is some redundant information in these features that has a weak association with the milling cutter wear. If all N features are directly input into the model, it will cause the model to process a large amount of features, resulting in a slow training speed and easy overfitting. Therefore, the prediction system calculates the importance score of each feature variable through the random forest algorithm (the higher the score, the stronger the influence on the wear amount), then sorts them from high to low, and selects the top Q (Q < N) core features and the corresponding wear amount at each moment to form a training dataset, which not only retains the key information valuable for prediction but also eliminates redundant interference, laying a foundation for the efficient training and accurate prediction of the subsequent neural network model.

[0071] Furthermore, the random forest consists of multiple independent decision trees. When training each decision tree, a part of the data is randomly sampled from the total samples for training (in-bag data), and the remaining unsampled data is the out-of-bag data. Since the out-of-bag data does not participate in the training of this decision tree, it can be used to test the prediction ability of the tree, and the calculated error is the out-of-bag error (b is the decision tree number). This error is the benchmark error when the feature is not disturbed, reflecting the accuracy of this decision tree in predicting the wear amount using the original features.

[0072] Next, for a certain feature variable to be evaluated , for all samples in the out-of-bag data add random noise (for example, randomly add or subtract a small value to the value of each sample to disrupt the original relationship with the wear amount), keep other features unchanged, and then input this set of features with added noise into this decision tree again, and calculate the prediction error at this time, which is the out-of-bag data error . If is an important feature for wear prediction, then after disrupting its relationship, the prediction accuracy of the decision tree will decrease significantly, will be larger than the benchmark error ; conversely, if is a redundant feature, the influence on the prediction result after adding noise is extremely small, and are basically close.

[0073] Finally, calculate the feature through the formulaThe importance score is calculated as the average of the differences between the baseline error and the interference error across all decision trees. A larger difference indicates a more significant contribution of the feature to prediction accuracy in most decision trees, resulting in a higher importance score. After sorting all N feature variables by importance score from highest to lowest, the top Q features are selected. These Q features at each time step are then paired with the corresponding measured wear amount (e.g., WT measured by a super-depth-of-field 3D microscope), thus forming the model training dataset.

[0074] S5. Train the neural network model according to the model training dataset to obtain the trained neural network model; based on the operating parameters corresponding to the target time of the milling cutter to be predicted, output the wear amount of the milling cutter to be predicted at the target time through the trained neural network model.

[0075] In one possible implementation, the neural network model is a deep neural network model, and the hidden layer of the neural network model is l layers; The output of the i-th neuron in the k-th hidden layer of the neural network model The calculation formula is: ; in, The number of neurons in the (k-1)th hidden layer is defined as The output of the j-th neuron in the (k-1)-th hidden layer is , It is the connection weight between the j-th neuron in the (k-1)-th hidden layer and the i-th neuron in the k-th hidden layer, and This represents the bias of the i-th neuron in the k-th hidden layer. This represents the tanh activation function, and its specific calculation formula is as follows: .

[0076] Furthermore, the output expression y of the neural network model is: ; Where b is the bias of the output layer neuron, and the variable is... This represents the output of the i-th neuron in the l-th hidden layer, while the variable... This represents the weight between the i-th neuron in the l-th hidden layer and the neuron in the output layer.

[0077] It should be noted that... As the data form of input variables for deep neural network models, Defined as the output of a deep neural network model, where m represents the number of input variables. Represented as the output variable of the k-th hidden layer, where, The number of neurons in the k-th hidden layer is defined as , As can be seen from S1-S5 above, the method provided by this embodiment of the invention constructs a basic dataset by using multi-time-time operating parameters and wear amount, which can ensure that the data covers the temporal process of milling cutter operation; the outlier removal and normalization processing in the preprocessing stage solves the problem of large noise in the original data and the impact of dimensional differences on model training, laying a high-quality data foundation for subsequent feature extraction; the feature extraction method combining CEEMDAN decomposition and statistical calculation can mine deep temporal features in non-stationary operating parameters, making up for the deficiency of traditional single feature extraction in capturing the subtle wear signals of milling cutters; the random forest algorithm selects highly important features, which can eliminate redundant information, reduce model complexity, and improve model training efficiency. Wear prediction is realized based on the trained neural network model, which can output accurate results by utilizing the complex correlation between deep features and wear amount, and finally realize online and accurate prediction of milling cutter wear, solving the problems of low efficiency and poor prediction accuracy of traditional manual stop detection, and ensuring the continuity of non-metallic idler shaft processing and product accuracy.

[0078] The foregoing mainly describes the solutions of the embodiments of the present invention from a methodological perspective. It is understood that, in order to achieve the above-mentioned functions, the prediction system 100 includes at least one of the hardware structures and software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of the present invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of the present invention.

[0079] In this embodiment of the invention, the prediction system 100 can be divided into functional units according to the above method example. For example, the prediction system 100 can be divided into functional units corresponding to various functions, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this embodiment of the invention is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0080] For example, Figure 4A schematic diagram of the hardware structure of a prediction system provided in an embodiment of the present invention is shown. The prediction system 100 includes: an acquisition module 110, used to acquire a basic dataset generated by a milling cutter during the machining of a non-metallic idler roller shaft, the basic dataset including the operating parameters and wear amount of the milling cutter at each of multiple time points; a preprocessing module 120, used to preprocess the basic dataset to obtain a preprocessed dataset, the preprocessing including outlier removal and normalization; a feature extraction module 130, used to extract deep feature information from the preprocessed dataset based on a temporal feature extraction algorithm combining CEEMDAN decomposition and statistical calculation, the deep feature information including N feature variables corresponding to each of multiple time points, each feature variable representing a different dimension of temporal features; and a training data generation module 1. 40 is used to determine the importance score of each feature variable using the random forest algorithm, and sort the N feature variables from largest to smallest based on the importance score of each feature variable to generate a model training dataset. The model training dataset includes the top Q feature variables and wear amount corresponding to each time step in multiple time steps. The importance score is positively correlated with the degree of influence on the wear amount of the milling cutter, Q is less than N, and Q and N are positive integers. The prediction module 150 is used to train the neural network model according to the model training dataset to obtain the trained neural network model. According to the input running parameters corresponding to the target time of the milling cutter to be predicted, the trained neural network model outputs the wear amount of the milling cutter to be predicted at the target time.

[0081] It should be understood that specific descriptions of the above-mentioned optional methods can be found in the foregoing method embodiments, and will not be repeated here. Furthermore, explanations of any of the prediction systems 100 provided above, as well as descriptions of their beneficial effects, can be found in the corresponding method embodiments described above, and will not be repeated here.

[0082] This invention also provides a computer-readable storage medium storing at least one computer instruction, which is loaded and executed by a processor to implement the methods of the various embodiments described above. Explanations of the relevant content and descriptions of the beneficial effects of any of the computer-readable storage media provided above can be found in the corresponding embodiments described above, and will not be repeated here.

[0083] This invention also provides a chip. This chip integrates a control circuit for implementing the functions of the prediction system 100 described above, and one or more ports. Optionally, the functions supported by this chip are as described above, and will not be repeated here.

[0084] Those skilled in the art will understand that the program for implementing all or part of the steps of the above embodiments, which can be executed by a program instructing related hardware, can be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a random access memory, etc. The processing unit or processor mentioned above can be a central processing unit, a general-purpose processor, an application-specific integrated circuit (ASIC), a microprocessor (DSP), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof.

[0085] This invention also provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform any of the methods described in the above embodiments. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this invention is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., SSD), etc.

[0086] It should be noted that the devices for storing computer instructions or computer programs provided in the embodiments of the present invention, such as, but not limited to, the aforementioned memory, computer-readable storage medium, and communication chip, are all non-transitory. Those skilled in the art should recognize that the functions described in the embodiments of the present invention in one or more of the above examples can be implemented using hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0087] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for predicting milling cutter wear in the machining of non-metallic idler roller shafts, characterized in that, The method includes: Acquire the basic dataset generated by the milling cutter during the machining of non-metallic idler roller shafts. The basic dataset includes the operating parameters and wear amount of the milling cutter at each time point at multiple times. The basic dataset is preprocessed to obtain a preprocessed dataset. The preprocessing includes removing outliers and normalization. A temporal feature extraction algorithm based on CEEMDAN decomposition and statistical calculation is used to extract deep feature information from the preprocessed dataset. The deep feature information includes N feature variables corresponding to each time step in multiple time steps, and each feature variable is used to characterize temporal features of different dimensions. The importance score of each feature variable is determined using the random forest algorithm, and the N feature variables are sorted from largest to smallest based on the importance score of each feature variable to generate a model training dataset. The model training dataset includes the top Q feature variables and wear amount at each time step in multiple time steps. The importance score is positively correlated with the degree of influence on the wear amount of the milling cutter, Q is less than N, and Q and N are positive integers. The neural network model is trained using the model training dataset to obtain a trained neural network model. Based on the operating parameters corresponding to the target time of the milling cutter, the trained neural network model outputs the wear amount of the milling cutter at the target time.

2. The method according to claim 1, characterized in that, The operating parameters include spindle speed, feed rate, depth of cut, width of cut, cutting force signals and vibration signals of the milling machine in the X, Y, and Z axes.

3. The method according to claim 2, characterized in that, The basic dataset generated by the milling cutter during the machining of non-metallic idler roller shafts includes: The spindle speed, feed rate, depth of cut, and width of cut are obtained at each moment in multiple time periods through the computer numerical control system of the milling machine. The cutting force and vibration signals of the milling machine in the X, Y, and Z axes are acquired by sensors. The wear amount at each time point in multiple time periods was obtained using a super depth-of-field 3D microscope; The basic dataset is generated based on the spindle speed, feed rate, depth of cut, width of cut, cutting force signals in the X, Y, and Z axes, vibration signals, and wear amount of the milling machine at each time point in multiple moments.

4. The method according to claim 3, characterized in that, The preprocessing of the basic dataset to obtain the preprocessed dataset includes: Determine the standard deviation and mean of each running parameter at each time step in the base dataset; Not located The operating parameters within the specified range were identified as outliers; Each outlier is replaced by the running parameters corresponding to the M time points before each outlier, resulting in a dataset after outlier removal, where M is a positive integer; Normalize each running parameter in the dataset after removing outliers to obtain a preprocessed dataset; The calculation formula for the normalization process is as follows: ； in, For the dataset after removing outliers, For the preprocessed dataset, and These are the minimum and maximum values ​​in the dataset after outliers have been removed.

5. The method according to claim 4, characterized in that, The temporal feature extraction algorithm based on CEEMDAN decomposition and statistical calculation extracts deep feature information from the preprocessed dataset, including: Based on a time-domain feature extraction algorithm combining CEEMDAN decomposition and statistical calculation, each operating parameter corresponding to each time step in the preprocessed dataset is decomposed to obtain t intrinsic mode functions and residuals corresponding to each operating parameter. Based on statistical calculation methods, multiple statistical parameters corresponding to each intrinsic mode function and residual are determined. These statistical parameters include sum, minimum, maximum, mean, standard deviation, variance, root mean square error, skewness, kurtosis, peak factor, and impulse factor. The deep feature information corresponding to each operating parameter is obtained by using the intrinsic mode function and multiple statistical parameters of the residual corresponding to each operating parameter at each time.

6. The method according to claim 5, characterized in that, The method of determining the importance score of each feature variable using the random forest algorithm includes: For characteristic variables The random forest algorithm is used to process each decision tree. Select the corresponding data outside the bag. Calculate the error outside the bag ,in, ; Random pair of feature variables Incorporate noise interference and calculate out-of-bag data error. ; Based on n bag outside errors Errors in data outside the bag Determine characteristic variables Importance score; feature Importance score The calculation formula is: 。 7. The method according to claim 6, characterized in that, The neural network model is a deep neural network model, and the hidden layer of the neural network model is l layers; The output of the i-th neuron in the k-th hidden layer of the neural network model The calculation formula is: ； in, The number of neurons in the (k-1)th hidden layer is defined as The output of the j-th neuron in the (k-1)-th hidden layer is , It is the connection weight between the j-th neuron in the (k-1)-th hidden layer and the i-th neuron in the k-th hidden layer, and This represents the bias of the i-th neuron in the k-th hidden layer. This represents the tanh activation function, and its specific calculation formula is as follows: 。 8. The method according to claim 7, characterized in that, The output expression y of the neural network model is: ； Where b is the bias of the output layer neuron, and the variable is... This represents the output of the i-th neuron in the l-th hidden layer, while the variable... This represents the weight between the i-th neuron in the l-th hidden layer and the neuron in the output layer.

9. A milling cutter wear prediction system for machining non-metallic idler roller shafts, characterized in that, The system includes: The acquisition module is used to acquire the basic dataset generated by the milling cutter during the machining of non-metallic idler roller shafts. The basic dataset includes the operating parameters and wear amount of the milling cutter at each time point in multiple moments. The preprocessing module is used to preprocess the basic dataset to obtain a preprocessed dataset. The preprocessing includes removing outliers and normalization. The feature extraction module is used to extract deep feature information from the preprocessed dataset using a temporal feature extraction algorithm that combines CEEMDAN decomposition and statistical calculation. The deep feature information includes N feature variables corresponding to each time step in multiple time steps, and each feature variable is used to characterize temporal features of different dimensions. The training data generation module is used to determine the importance score of each feature variable using the random forest algorithm, and to sort the N feature variables from largest to smallest based on the importance score of each feature variable to generate a model training dataset. The model training dataset includes the top Q feature variables and wear amount at each time step in multiple time steps. The importance score is positively correlated with the degree of influence on the wear amount of the milling cutter, Q is less than N, and Q and N are positive integers. The prediction module is used to train the neural network model based on the model training dataset to obtain the trained neural network model; based on the input operating parameters corresponding to the target time of the milling cutter to be predicted, the trained neural network model outputs the wear amount of the milling cutter to be predicted at the target time.

10. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the milling cutter wear prediction method for non-metallic roller shaft machining as described in any one of claims 1-8.

Images