Longitudinal shear tool on-line sharpening decision method based on acoustic vibration temperature multi-modal perception

Abstract

The present application relates to the technical field of slitting equipment cutter state monitoring and maintenance, and discloses a slitting cutter online sharpening decision method based on acoustic vibration temperature multi-modal sensing, which is executed on a slitting equipment configured with an acoustic emission sensor, a vibration sensor, a temperature sensor, a sharpening execution mechanism and a data processing unit, and the data processing unit executes the following steps: S1, in the strip shearing operation process of the slitting equipment, the original acoustic emission signal of the slitting cutter is collected through the acoustic emission sensor, the original vibration signal of the slitting cutter is collected through the vibration sensor, and the original temperature signal of the slitting cutter is collected through the temperature sensor. The present application has the advantages of multi-modal fusion, online decision, accurate sharpening parameters, reduction of manual intervention and the like.

Description

Technical Field

[0001] This invention relates to the field of slitting equipment tool condition monitoring and maintenance technology, specifically to an online tool sharpening decision-making method for slitting tools based on multimodal sensing of acoustic vibration and temperature. Background Technology

[0002] In the strip shearing process of slitting equipment, the cutting edge condition of the slitting tools directly affects the strip cut quality, production efficiency, and equipment operation stability. As slitting equipment develops towards high speed and continuous operation, the blades will gradually wear, chip, or become dull during long-term shearing operations. If timely sharpening and maintenance are not performed, it will lead to increased burrs on the strip cut, decreased dimensional accuracy, and even failures such as strip breakage and machine shutdown.

[0003] Currently, the decision-making process for sharpening scissors largely relies on operator experience or single signal monitoring. For example, the tool condition is judged solely by changes in vibration signal amplitude, or the degree of wear is inferred solely by the degree of temperature rise. However, the information reflected by a single signal is limited and easily affected by fluctuations in operating conditions, environmental noise, and material differences, making it difficult to accurately distinguish between normal and abnormal tool wear, and prone to misjudgment. Furthermore, existing manual decision-making methods lack unified standards; parameters such as sharpening feed rate, sharpening duration, and sharpening reciprocation number are often set based on experience, easily leading to over-sharpening or under-sharpening. Over-sharpening shortens tool life and increases costs; under-sharpening fails to restore cutting edge sharpness, affecting subsequent shearing quality. In addition, current technologies struggle to achieve online decision-making during shearing operations, typically requiring machine shutdown for inspection or offline testing, impacting continuous production efficiency.

[0004] Therefore, how to design a method that can integrate multimodal information of acoustic emission, vibration and temperature, and determine the tool status online and automatically output a sharpening decision during the operation of the slitting equipment has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide an online grinding decision method for scissor cutters based on multimodal sensing of acoustic vibration and temperature. This method has advantages such as multimodal fusion sensing, online real-time decision-making, accurate grinding parameters, reduced manual intervention, and improved tool life, thus solving the problems in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for online sharpening of slitting shears based on multimodal acoustic, vibration, and temperature sensing is implemented on a slitting machine equipped with an acoustic emission sensor, a vibration sensor, a temperature sensor, a sharpening actuator, and a data processing unit. The method is characterized by the data processing unit performing the following steps: S1. During the strip shearing operation of the slitting equipment, the original acoustic emission signal of the slitting tool is collected by an acoustic emission sensor, the original vibration signal of the slitting tool is collected by a vibration sensor, and the original temperature signal of the slitting tool is collected by a temperature sensor. S2. Perform digital filtering, abnormal segment removal, and time axis alignment on the original acoustic emission signal, original vibration signal, and original temperature signal respectively to obtain the time-axis aligned acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence. S3. Extract time-frequency domain features from the acoustic emission preprocessing sequence and the vibration preprocessing sequence respectively, and extract time-series statistical features from the temperature preprocessing sequence to obtain the original multimodal feature set; S4. Perform feature normalization on the original multimodal feature set to obtain normalized multimodal feature vectors; S5. Input the normalized multimodal feature vector into the pre-constructed and offline trained decision model, and output the discretized sharpening decision code; S6. Read the preset parameter group corresponding to the grinding decision code and output an electrical signal to the grinding actuator. The preset parameter group includes the grinding feed amount, grinding duration and grinding reciprocation number.

[0007] Preferably, in the acquisition process of step S1, the acoustic emission sensor used is a piezoelectric acoustic emission sensor with a resonant frequency of 100kHz to 400kHz, which is fixedly installed on the end face of the cutter shaft or on the bearing seat of the slitting tool using a coupling agent; the vibration sensor used is a piezoelectric accelerometer sensor, which is fixedly installed in the radial direction of the bearing seat of the slitting tool; the temperature sensor used is an infrared temperature sensor, whose detection direction is aligned with the cutting area of ​​the cutting edge of the slitting tool; the acoustic emission sensor, vibration sensor and temperature sensor are synchronously started to acquire data through the same data acquisition card with the same trigger signal.

[0008] It is worth noting that by using the same data acquisition card and starting the acquisition synchronously with the same trigger signal, the consistency of acoustic emission signals, vibration signals and temperature signals on the time reference is ensured, avoiding the time deviation of multimodal data caused by asynchronous acquisition, and providing a reliable data foundation for subsequent time axis alignment and feature fusion.

[0009] Preferably, in step S2, the digital filtering process includes: applying a bandpass filter of 100kHz to 1MHz to the original acoustic emission signal, applying a bandpass filter of 1Hz to 20kHz to the original vibration signal, and applying a first-order low-pass filter to the original temperature signal.

[0010] It is worth noting that setting different filtering frequency bands for acoustic emission signals, vibration signals, and temperature signals can effectively filter out power frequency interference, environmental noise, and high-frequency clutter in each modal signal, retain the effective frequency components related to tool wear state, and improve the signal-to-noise ratio of subsequent feature extraction.

[0011] Preferably, in step S2, the abnormal segment removal process includes: setting threshold values ​​for acoustic emission signal amplitude, vibration signal amplitude, and temperature signal amplitude, respectively, and removing signal segments that exceed the corresponding amplitude thresholds from the original acoustic emission signal, original vibration signal, and original temperature signal.

[0012] It is worth noting that by setting an amplitude threshold to remove abnormal segments, large-amplitude abnormal signals caused by strip joints, foreign object impacts, or instantaneous sensor interference can be prevented from entering the subsequent processing flow, thus preventing misjudgments caused by these non-tool wear factors and improving the quality of input data for the decision-making model.

[0013] Preferably, in step S2, the time axis alignment process includes: using the sampling time step corresponding to the lowest sampling frequency in the acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence as the unified sampling time step, and resampling the acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence to the unified sampling time step.

[0014] It is worth noting that using the sampling time step corresponding to the lowest sampling frequency as the unified sampling time step can unify multimodal signals with different sampling rates to the same time dimension without losing key information. This facilitates subsequent feature splicing and decision model processing, and ensures the one-to-one correspondence of multimodal features in time.

[0015] Preferably, in step S3, continuous wavelet transform is performed on the acoustic emission preprocessing sequence and the vibration preprocessing sequence respectively, and the energy mean and energy variance of the wavelet coefficients are extracted as time-frequency domain features; the temperature preprocessing sequence is calculated using a sliding window to calculate the temperature mean and temperature change slope within the window as time-series statistical features.

[0016] It is worth noting that continuous wavelet transform can simultaneously characterize the local features of acoustic emission signals and vibration signals in both the time and frequency domains. Extracting the mean energy and energy variance can effectively reflect the changes in energy distribution during tool wear. The mean temperature and temperature change slope can reflect the trend of thermal state changes in the cutting edge shearing region. Together, these three constitute a multi-dimensional description of the tool wear state.

[0017] Preferably, in step S4, the feature normalization process adopts max-min normalization, which maps each feature in the original multimodal feature set to the [0,1] value range, and concatenates the mapped features into a one-dimensional normalized multimodal feature vector in a preset order.

[0018] It is worth noting that max-min normalization can eliminate the influence of differences in the scale and numerical range of different features on the decision model, so that each modality feature can participate in training and decision-making at the same scale; it can be concatenated into a one-dimensional vector in a preset order, which is convenient for direct input into a multi-layer fully connected neural network and simplifies the data processing process.

[0019] Preferably, in step S5, the decision model is a multi-layer fully connected neural network, including an input layer, at least two hidden layers, and an output layer. The number of neurons in each hidden layer decreases progressively, and adjacent layers are connected by a fully connected weight matrix. An activation function is configured after each layer. The offline training of the decision model is a pre-construction process before the execution of this method. The offline training includes: obtaining a labeled historical sample dataset, where the labels of the historical sample dataset are discretized sharpening decision codes corresponding to the sharpening action; dividing the historical sample dataset into a training subset and a validation subset; performing iterative training on the multi-layer fully connected neural network using the cross-entropy loss function; stopping the iteration when the recognition accuracy of the validation subset reaches a preset accuracy threshold, or when the recognition accuracy of the validation subset no longer increases in a consecutive preset number of iterations; and saving the trained network weight parameters to obtain the decision model.

[0020] It is worth noting that using a multi-layer fully connected neural network with a progressively decreasing hidden layer structure can control model complexity while ensuring classification ability; the cross-entropy loss function is suitable for classification tasks with discretized sharpening decision codes; by validating subset accuracy thresholds and early stopping conditions, overfitting can be avoided, and the generalization ability and stability of the decision model during online operation can be improved.

[0021] Preferably, the discretized grinding decision code includes a first code, a second code, and a third code; the first code corresponds to a preset parameter set where the grinding feed is zero, the grinding duration is zero, and the grinding reciprocation count is zero; the second code corresponds to a preset parameter set where the grinding feed is a first preset value, the grinding duration is a second preset value, and the grinding reciprocation count is a third preset value; the third code corresponds to a preset parameter set where the grinding feed is a fourth preset value, the grinding duration is a fifth preset value, and the grinding reciprocation count is a sixth preset value; the fourth preset value is greater than the first preset value, the fifth preset value is greater than the second preset value, and the sixth preset value is greater than the third preset value.

[0022] It is worth noting that the sharpening decision is discretized into a first code, a second code, and a third code, corresponding to three decision states: no sharpening, light sharpening, and heavy sharpening, respectively. This allows the output of the decision model to directly correspond to the preset parameter set of the sharpening actuator, eliminating the need for manual secondary calculations and improving the response speed and execution consistency of online decision-making.

[0023] Preferably, the process further includes step S7: After the grinding actuator completes the physical grinding action corresponding to the preset parameter group, it collects a new round of original acoustic emission signals, original vibration signals and original temperature signals through acoustic emission sensors, vibration sensors and temperature sensors, and repeats steps S2 to S6 until the newly output grinding decision code is the first code, then the loop ends.

[0024] It is worth noting that after the grinding action is completed by the grinding actuator in step S7, the multimodal signal is collected again and the decision-making process is repeated to form a closed-loop control until the decision model outputs the first code, i.e., the no-grind state, thereby confirming that the tool has been restored to the normal state, avoiding over-grinding or under-grinding, and improving the accuracy and automation of grinding maintenance.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention synchronously acquires multimodal signals of the slitting tool using acoustic emission sensors, vibration sensors, and temperature sensors, and then processes them through digital filtering, abnormal segment removal, and time axis alignment. This solves the problems of limited information from single signal monitoring and susceptibility to interference from working conditions, and significantly improves the accuracy and reliability of tool wear condition judgment. 2. This invention extracts the time-frequency domain features of acoustic emission and vibration signals and the time-series statistical features of temperature signals, and performs normalization and feature splicing to construct a multimodal feature vector that can comprehensively reflect the tool wear state, providing rich and standardized input data for the decision model; 3. This invention outputs discrete grinding decision codes by pre-constructing and offline-training a multi-layer fully connected neural network, and reads the preset parameter sets corresponding to the codes, such as grinding feed, grinding duration, and grinding reciprocation number, to realize online and automated grinding decision-making for longitudinal scissors, reducing reliance on human experience and improving the consistency of grinding parameters and execution efficiency. 4. In step S7, the present invention collects multimodal signals again after the grinding action and makes cyclic decisions until the first code is output, forming a closed-loop control, which effectively avoids over-grinding and under-grinding, extends the service life of the slitting tool, and reduces production and maintenance costs. Attached Figure Description

[0026] Figure 1 The flowchart shown is a flowchart of the online sharpening decision method for longitudinal shear tools based on multimodal sensing of acoustic vibration and temperature according to the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] To address the limitations of existing technologies, such as limited information from single-signal monitoring, reliance on manual experience for grinding decisions, susceptibility to over-grinding or under-grinding, and difficulty in real-time online assessment, the following technical solution is proposed. Please refer to [link / reference needed]. Figure 1 .

[0029] A method for online sharpening of slitting cutters based on multimodal sensing of acoustic emission, vibration, and temperature is proposed. The method is executed on a slitting device equipped with an acoustic emission sensor, a vibration sensor, a temperature sensor, a sharpening actuator, and a data processing unit. The data processing unit performs the following steps: S1. During the strip shearing operation of the slitting equipment, the original acoustic emission signal of the slitting tool is collected by an acoustic emission sensor, the original vibration signal of the slitting tool is collected by a vibration sensor, and the original temperature signal of the slitting tool is collected by a temperature sensor. S2. Perform digital filtering, abnormal segment removal, and time axis alignment on the original acoustic emission signal, original vibration signal, and original temperature signal respectively to obtain the time-axis aligned acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence. S3. Extract time-frequency domain features from the acoustic emission preprocessing sequence and the vibration preprocessing sequence respectively, and extract time-series statistical features from the temperature preprocessing sequence to obtain the original multimodal feature set; S4. Perform feature normalization on the original multimodal feature set to obtain normalized multimodal feature vectors; S5. Input the normalized multimodal feature vector into the pre-constructed and offline trained decision model, and output the discretized sharpening decision code; S6. Read the preset parameter group corresponding to the grinding decision code and output an electrical signal to the grinding actuator. The preset parameter group includes the grinding feed amount, grinding duration and grinding reciprocation number.

[0030] In this embodiment, specifically, during the acquisition process in step S1, the acoustic emission sensor used is a piezoelectric acoustic emission sensor with a resonant frequency of 100kHz to 400kHz, which is fixedly installed on the end face of the cutter shaft or on the bearing seat of the slitting tool using a coupling agent; the vibration sensor used is a piezoelectric accelerometer sensor, which is fixedly installed in the radial direction of the bearing seat of the slitting tool; the temperature sensor used is an infrared temperature sensor, whose detection direction is aligned with the cutting area of ​​the cutting edge of the slitting tool; the acoustic emission sensor, vibration sensor and temperature sensor are synchronously started to acquire data through the same data acquisition card with the same trigger signal.

[0031] In this embodiment, specifically, in step S2, the digital filtering process includes: applying a bandpass filter of 100kHz to 1MHz to the original acoustic emission signal, applying a bandpass filter of 1Hz to 20kHz to the original vibration signal, and applying a first-order low-pass filter to the original temperature signal.

[0032] In this embodiment, specifically, in step S2, the abnormal segment removal process includes: setting an amplitude threshold for acoustic emission signal, an amplitude threshold for vibration signal, and an amplitude threshold for temperature signal, respectively, and removing signal segments that exceed the corresponding amplitude thresholds from the original acoustic emission signal, the original vibration signal, and the original temperature signal.

[0033] In this embodiment, specifically, in step S2, the time axis alignment process includes: using the sampling time step corresponding to the lowest sampling frequency in the acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence as the unified sampling time step, and resampling the acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence to the unified sampling time step.

[0034] In this embodiment, specifically, in step S3, continuous wavelet transform is performed on the acoustic emission preprocessing sequence and the vibration preprocessing sequence respectively, and the energy mean and energy variance of the wavelet coefficients are extracted as time-frequency domain features; the temperature preprocessing sequence is calculated using a sliding window to calculate the temperature mean and temperature change slope within the window as time-series statistical features.

[0035] In this embodiment, specifically, in step S4, the feature normalization process adopts max-min normalization, which maps each feature in the original multimodal feature set to the [0,1] value range, and concatenates the mapped features into a one-dimensional normalized multimodal feature vector in a preset order.

[0036] In this embodiment, specifically, in step S5, the decision model is a multi-layer fully connected neural network, including an input layer, at least two hidden layers, and an output layer. The number of neurons in each hidden layer decreases progressively, and adjacent layers are connected by a fully connected weight matrix. An activation function is configured after each layer. The offline training of the decision model is a pre-construction process before the execution of this method. The offline training includes: obtaining a labeled historical sample dataset, the labels of which are discretized sharpening decision codes corresponding to the sharpening action; dividing the historical sample dataset into a training subset and a validation subset; performing iterative training on the multi-layer fully connected neural network using the cross-entropy loss function; stopping the iteration when the recognition accuracy of the validation subset reaches a preset accuracy threshold, or when the recognition accuracy of the validation subset no longer increases in a preset number of iterations; and saving the trained network weight parameters to obtain the decision model.

[0037] In this embodiment, specifically, the discretized grinding decision code includes a first code, a second code, and a third code; the first code corresponds to a preset parameter set where the grinding feed is zero, the grinding duration is zero, and the grinding reciprocation count is zero; the second code corresponds to a preset parameter set where the grinding feed is a first preset value, the grinding duration is a second preset value, and the grinding reciprocation count is a third preset value; the third code corresponds to a preset parameter set where the grinding feed is a fourth preset value, the grinding duration is a fifth preset value, and the grinding reciprocation count is a sixth preset value; the fourth preset value is greater than the first preset value, the fifth preset value is greater than the second preset value, and the sixth preset value is greater than the third preset value.

[0038] In this embodiment, specifically, it also includes step S7: After the grinding actuator completes the physical grinding action corresponding to the preset parameter group, it collects a new round of original acoustic emission signal, original vibration signal and original temperature signal through acoustic emission sensor, vibration sensor and temperature sensor, and repeats steps S2 to S6 until the newly output grinding decision code is the first code, then the loop ends.

[0039] Working Principle: During operation, the slitting equipment is in strip shearing mode. The data processing unit first executes step S1, acquiring the original acoustic emission signal of the slitting tool through an acoustic emission sensor, the original vibration signal of the slitting tool through a vibration sensor, and the original temperature signal of the slitting tool through a temperature sensor. Then, step S2 is executed, performing digital filtering, outlier removal, and time-axis alignment on the three original signals respectively, resulting in time-aligned acoustic emission preprocessing sequences, vibration preprocessing sequences, and temperature preprocessing sequences. Next, step S3 is executed, extracting time-frequency domain features such as wavelet coefficient energy mean and energy variance from the acoustic emission and vibration preprocessing sequences, and extracting time-series statistical features such as sliding window temperature mean and temperature change slope from the temperature preprocessing sequence, forming a multimodal original feature set. S4: Perform max-min normalization on the original multimodal feature set and concatenate it into a one-dimensional normalized multimodal feature vector according to a preset order; Execute step S5: Input the normalized multimodal feature vector into a pre-constructed and offline-trained multilayer fully connected neural network decision model, and output a discrete grinding decision code; Execute step S6: Read the preset parameter set corresponding to the code, including the grinding feed, grinding duration, and grinding reciprocation number, and output an electrical signal to the grinding actuator, which then performs the corresponding physical grinding action; After the grinding action is completed, execute step S7: Collect a new round of original acoustic emission signals, original vibration signals, and original temperature signals, and repeat steps S2 to S6 until the newly output grinding decision code is the first code, indicating that the tool has returned to normal, and end this online grinding decision cycle.

[0040] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0041] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for online sharpening of slitting shears based on multimodal acoustic, vibration, and temperature sensing, wherein the method is executed on a slitting device equipped with an acoustic emission sensor, a vibration sensor, a temperature sensor, a sharpening actuator, and a data processing unit, characterized in that, The data processing unit performs the following steps: S1. During the strip shearing operation of the slitting equipment, the original acoustic emission signal of the slitting tool is collected by an acoustic emission sensor, the original vibration signal of the slitting tool is collected by a vibration sensor, and the original temperature signal of the slitting tool is collected by a temperature sensor. S2. Perform digital filtering, abnormal segment removal, and time axis alignment on the original acoustic emission signal, original vibration signal, and original temperature signal respectively to obtain the time-axis aligned acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence. S3. Extract time-frequency domain features from the acoustic emission preprocessing sequence and the vibration preprocessing sequence, respectively, and extract time-series statistical features from the temperature preprocessing sequence to obtain the original multimodal feature set; S4. Perform feature normalization on the original multimodal feature set to obtain normalized multimodal feature vectors; S5. Input the normalized multimodal feature vector into the pre-constructed and offline trained decision model, and output the discretized sharpening decision code; S6. Read the preset parameter group corresponding to the grinding decision code and output an electrical signal to the grinding actuator. The preset parameter group includes the grinding feed amount, grinding duration and grinding reciprocation number.

2. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, During the data acquisition process in step S1, the acoustic emission sensor used is a piezoelectric acoustic emission sensor with a resonant frequency of 100kHz to 400kHz, which is fixedly installed on the end face of the cutter shaft or on the bearing seat of the slitting tool using a coupling agent; the vibration sensor used is a piezoelectric accelerometer sensor, which is fixedly installed in the radial direction of the bearing seat of the slitting tool; the temperature sensor used is an infrared temperature sensor, whose detection direction is aligned with the cutting area of ​​the cutting edge of the slitting tool; the acoustic emission sensor, vibration sensor and temperature sensor are synchronously started to acquire data through the same data acquisition card with the same trigger signal.

3. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, In step S2, the digital filtering process includes: applying a bandpass filter of 100kHz to 1MHz to the original acoustic emission signal, applying a bandpass filter of 1Hz to 20kHz to the original vibration signal, and applying a first-order low-pass filter to the original temperature signal.

4. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, In step S2, the abnormal segment removal process includes: setting threshold values ​​for acoustic emission signal amplitude, vibration signal amplitude, and temperature signal amplitude, respectively, and removing signal segments that exceed the corresponding amplitude thresholds from the original acoustic emission signal, original vibration signal, and original temperature signal.

5. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, In step S2, the time axis alignment process includes: using the sampling time step corresponding to the lowest sampling frequency in the acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence as the unified sampling time step, and resampling the acoustic emission preprocessing sequence, vibration preprocessing sequence, and temperature preprocessing sequence to the unified sampling time step.

6. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, In step S3, continuous wavelet transform is performed on the acoustic emission preprocessing sequence and the vibration preprocessing sequence respectively, and the energy mean and energy variance of the wavelet coefficients are extracted as time-frequency domain features. For the temperature preprocessing sequence, a sliding window is used to calculate the mean temperature and the slope of temperature change within the window as time series statistical features.

7. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, In step S4, the feature normalization process adopts max-min normalization, which maps each feature in the original multimodal feature set to the [0,1] value range, and then concatenates the mapped features into a one-dimensional normalized multimodal feature vector in a preset order.

8. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 1, characterized in that, In step S5, the decision model is a multi-layer fully connected neural network, including an input layer, at least two hidden layers, and an output layer. The number of neurons in each hidden layer decreases progressively, and adjacent layers are connected by a fully connected weight matrix. An activation function is configured after each layer. The offline training of the decision model is a pre-construction process before the execution of this method. Offline training includes: obtaining a labeled historical sample dataset, where the labels of the historical sample dataset are discretized sharpening decision codes corresponding to the sharpening action; dividing the historical sample dataset into a training subset and a validation subset; performing iterative training on the multi-layer fully connected neural network using the cross-entropy loss function; stopping the iteration when the recognition accuracy of the validation subset reaches a preset accuracy threshold, or when the recognition accuracy of the validation subset no longer increases in consecutive preset iterations; and saving the trained network weight parameters to obtain the decision model.

9. The online sharpening decision method for longitudinal shear tools based on acoustic, vibration, and temperature multimodal sensing according to claim 8, characterized in that, The discretized grinding decision code includes a first code, a second code, and a third code. The first code corresponds to a preset parameter set where the grinding feed is zero, the grinding duration is zero, and the grinding reciprocation count is zero. The second code corresponds to a preset parameter set where the grinding feed is a first preset value, the grinding duration is a second preset value, and the grinding reciprocation count is a third preset value. The third code corresponds to a preset parameter set where the grinding feed is a fourth preset value, the grinding duration is a fifth preset value, and the grinding reciprocation count is a sixth preset value. The fourth preset value is greater than the first preset value, the fifth preset value is greater than the second preset value, and the sixth preset value is greater than the third preset value.

10. The online sharpening decision method for longitudinal shear tools based on multimodal acoustic, vibration, and temperature sensing according to claim 9, characterized in that, It also includes step S7: After the grinding actuator completes the physical grinding action corresponding to the preset parameter group, it collects a new round of original acoustic emission signal, original vibration signal and original temperature signal through acoustic emission sensor, vibration sensor and temperature sensor, and repeats steps S2 to S6 until the newly output grinding decision code is the first code, then the loop ends.

Images