A method for predicting the lifespan of kitchen garbage disposer blades using deep learning algorithms

Abstract

The application discloses a kind of utilize deep learning algorithm to predict kitchen garbage disposer cutter head life method, belong to data processing technical field, comprising: obtain the working condition parameter of kitchen garbage disposer, generate the standardized data set containing working condition parameter;The vibration signal spectrum feature and motor current harmonic feature of standardized data set are extracted, and multi-dimensional feature set is constructed;Multi-dimensional feature set is input into pre-trained long short-term memory network model, and the evaluation value of cutter head wear degree is output;The vibration signal spectrum feature and motor current harmonic feature corresponding to the evaluation value of cutter head wear degree that exceeds preset threshold are extracted, according to multi-dimensional feature set matching history record, using sliding window weighted algorithm calculates residual life prediction value.The kitchen garbage disposer cutter head life method utilizing deep learning algorithm solves the problem that current cutter head life is difficult to predict by the correlation between working condition parameter and residual life, thereby leading to the problem of low cutter head life prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for predicting the lifespan of kitchen waste disposer blades using deep learning algorithms. Background Technology

[0002] In the field of smart home appliances, kitchen waste disposers are crucial for improving quality of life, and their lifespan directly impacts user experience and equipment safety. With the widespread adoption of smart technologies, predictive maintenance through data-driven approaches has become a critical issue in this field. One of the core components of a kitchen waste disposer—the blade—has its wear directly determining the device's efficiency and lifespan; therefore, accurate prediction of blade lifespan is paramount. However, current methods for predicting blade lifespan largely rely on traditional periodic inspections or simple threshold judgments. These methods often fail to accurately capture the dynamic changes in blade wear and are ill-suited to the complex operating conditions of different usage scenarios, resulting in inaccurate predictions. This can lead to the device operating with defects unnoticed, or even posing safety hazards. Therefore, current methods for predicting kitchen waste disposer blade lifespan struggle to establish a correlation between operating parameters and remaining lifespan, resulting in low accuracy in blade lifespan prediction. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides a method for predicting the lifespan of kitchen waste disposer blades using deep learning algorithms, thereby solving the aforementioned problems.

[0004] The technical solution adopted by this invention to solve its technical problem is: a method for predicting the lifespan of a kitchen garbage disposer blade using a deep learning algorithm, comprising the following steps:

[0005] S1: Obtain the real-time collected operating parameters of the kitchen waste disposer and generate a standardized dataset containing the operating parameters, including motor operating status data and blade vibration frequency data.

[0006] S2: Perform time-frequency joint analysis on the standardized dataset to extract the spectral features of vibration signals and the harmonic features of motor current, and construct a multi-dimensional feature set;

[0007] S3: Input the multidimensional feature set containing the spectral characteristics of vibration signals and the harmonic characteristics of motor current into the pre-trained long short-term memory network model, and output the evaluation value of the wear degree of the tool head;

[0008] S4: Extract the vibration signal spectrum features and motor current harmonic features corresponding to the evaluation value of the tool wear degree exceeding the preset threshold. Based on the multi-dimensional feature set matching historical records, calculate the remaining life prediction value using a sliding window weighted algorithm.

[0009] Specifically, in step S1, after obtaining the operating parameters, noise elimination and outlier processing are performed on the operating parameters to generate a standardized dataset including timestamp-aligned motor operating status data and cutter head vibration frequency data.

[0010] Preferably, in step S2, a short-time Fourier transform is used to perform time-frequency analysis on the tool vibration frequency data in the standardized dataset, the window function is set to a Hanning window, the window length is a preset sampling point, and the time-frequency distribution matrix is ​​obtained.

[0011] For the time-frequency distribution matrix, divide it into multiple frequency bands for energy integration, and calculate the energy proportion of each frequency band as preliminary spectral characteristics;

[0012] From the preliminary spectral features, wavelet decomposition technology is used to extract the core vibration feature frequency bands, and the average value and variance of the core vibration feature frequency bands are calculated to form the spectral features of the vibration signal.

[0013] Optionally, in step S2, a fast Fourier transform is performed on the motor current signal in the motor operating status data of the standardized dataset to analyze multiple harmonic components, calculate the amplitude ratio of each harmonic, and obtain the motor current harmonic characteristics.

[0014] It is worth noting that in step S2, based on the vibration signal spectrum features and the motor current harmonic features, a multidimensional feature set is generated by aligning the timestamps and using a horizontal splicing method. The multidimensional feature set includes the vibration signal spectrum features and the motor current harmonic features.

[0015] Specifically, in step S3, the multidimensional feature set is processed using a standardization method to obtain a standardized feature dataset;

[0016] The standardized feature dataset is input into a pre-trained long short-term memory network model, and time-series analysis is performed on the standardized feature dataset to obtain an evaluation value of the tool wear degree.

[0017] Preferably, in step S4, the step of matching historical maintenance records according to the multidimensional feature set includes: extracting the vibration signal spectrum features and motor current harmonic features corresponding to the evaluation value of the tool wear degree exceeding the preset threshold from the multidimensional feature set, and generating an abnormal feature dataset;

[0018] Historical records are retrieved from a pre-established database, which includes historical data. A similarity matching method is used to determine historical data that matches the abnormal feature dataset as a subset of historical data. The subset of historical data includes historical vibration signal spectral features and historical motor current harmonic features.

[0019] It is worth noting that in step S4, the step of calculating the remaining lifetime prediction interval using the sliding window weighted algorithm includes: for the matched historical data subset, the features in the historical data are segmented using a sliding window method to obtain a segmented feature sequence;

[0020] Based on the segmented feature sequence, a weighted method is used to calculate the weighted feature value corresponding to each segmented feature sequence according to the weights of the historical vibration signal spectrum features and historical motor current harmonic features in the historical data subset.

[0021] The weighted feature values ​​and the corresponding timestamp data of the segmented feature sequences are processed using time series analysis methods to obtain feature change trend parameters;

[0022] The historical records also include historical maintenance records. The weighted feature value at the time of tool head replacement in the historical maintenance records is obtained. The time required for the current weighted feature value to reach the weighted feature value at the time of tool head replacement in the historical maintenance records is calculated based on the feature change trend parameter, and used as the remaining life prediction value.

[0023] The beneficial effects of this invention are as follows: In the method for predicting the blade life of a kitchen waste disposer using a deep learning algorithm, the operating parameters of the disposer, including motor operating status and blade vibration frequency data, are collected in real time. Vibration signal spectral features and motor current harmonic features are extracted to construct a multi-dimensional feature set. This multi-dimensional feature set is input into a pre-trained long short-term memory network model, which outputs an evaluation value of the blade wear degree. For wear evaluation values ​​exceeding a threshold, combined with historical maintenance records, a sliding window weighted algorithm is used to predict the remaining life. This invention, through multi-dimensional data analysis and deep learning methods, predicts blade life by linking operating parameters with remaining life, achieving accurate assessment of the wear state of kitchen waste disposer blades and reliable prediction of their remaining life. This facilitates timely maintenance and replacement, improving equipment operating efficiency and safety. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a method for predicting the lifespan of a kitchen waste disposer blade using a deep learning algorithm, as described in one embodiment of the present invention.

[0025] Figure 2 This is a flowchart illustrating the steps of step S2 in one embodiment of the present invention;

[0026] Figure 3 This is a flowchart illustrating step S4 in one embodiment of the present invention. Detailed Implementation

[0027] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0028] like Figure 1-3 As shown, a method for predicting the lifespan of a kitchen garbage disposer blade using a deep learning algorithm includes the following steps:

[0029] S1: Obtain the real-time collected operating parameters of the kitchen waste disposer and generate a standardized dataset containing the operating parameters, including motor operating status data and blade vibration frequency data.

[0030] S2: Perform time-frequency joint analysis on the standardized dataset to extract the spectral features of vibration signals and the harmonic features of motor current, and construct a multi-dimensional feature set;

[0031] S3: Input the multidimensional feature set containing the spectral characteristics of vibration signals and the harmonic characteristics of motor current into the pre-trained long short-term memory network model, and output the evaluation value of the wear degree of the tool head;

[0032] S4: Extract the vibration signal spectrum features and motor current harmonic features corresponding to the evaluation value of the tool wear degree exceeding the preset threshold. Based on the multi-dimensional feature set matching historical records, calculate the remaining life prediction value using a sliding window weighted algorithm.

[0033] In the method for predicting the blade life of a kitchen waste disposer using a deep learning algorithm, real-time acquisition of the disposer's operating parameters, including motor operating status and blade vibration frequency data, is used to extract vibration signal spectral features and motor current harmonic features, constructing a multi-dimensional feature set. This multi-dimensional feature set is input into a pre-trained long short-term memory network model, which outputs an assessment value of the blade wear degree. For wear assessment values ​​exceeding a threshold, combined with historical maintenance records, a sliding window weighted algorithm is used to predict the remaining lifespan. This invention, through multi-dimensional data analysis and deep learning methods, predicts blade lifespan by linking operating parameters with remaining lifespan, achieving accurate assessment of the wear state of kitchen waste disposer blades and reliable prediction of their remaining lifespan. This facilitates timely maintenance and replacement, improving equipment operating efficiency and safety.

[0034] It is worth noting that in step S1, after obtaining the operating parameters, noise elimination and outlier processing are performed on the operating parameters to generate a standardized dataset including timestamp-aligned motor operating status data and cutter head vibration frequency data.

[0035] In this embodiment, a raw dataset of real-time operating parameters is acquired from the kitchen waste disposer via a sensor network. This raw dataset contains multiple time-series sampling points, resulting in timestamp-aligned motor operating status data and blade vibration frequency data. Timestamp alignment ensures the correspondence between the two types of data at the same point in time, providing a prerequisite for subsequent feature fusion. A sliding window method is used to smooth the raw dataset, generating a denoised first intermediate dataset. For the first intermediate dataset, a dynamic threshold range is calculated for each sampling point. This dynamic threshold range is based on the interquartile range, determining the upper and lower thresholds. If a sampling point in the first intermediate dataset exceeds the dynamic threshold range, that sampling point is discarded, generating a second intermediate dataset with null defects. Null defects in the second intermediate dataset are filled using a linear interpolation method to obtain a complete, standardized dataset.

[0036] For example, raw datasets obtained from kitchen waste disposers via sensor networks often contain multiple time-series sampling points, reflecting the operational status of the disposer at different times. These raw datasets may be affected by noise, such as sudden vibrations during equipment operation or outlier values ​​collected by sensors, thus requiring smoothing. The sliding window method is a commonly used denoising technique. Assuming a window size of 5 sampling points, it smooths the data fluctuations at each point by calculating the average value within the window. For instance, if the blade vibration frequency data collected within a certain time period are 10Hz, 12Hz, 50Hz, 11Hz, and 9Hz, then 50Hz is clearly an outlier. After smoothing with a sliding window, the value at this point may be adjusted to be close to the average value of the surrounding data, such as 10.4Hz, thus generating the first intermediate dataset.

[0037] Specifically, calculating the dynamic threshold range is a crucial step for the first intermediate dataset. The dynamic threshold range is based on the interquartile range to determine the upper and lower thresholds, adapting to changes in data distribution. For example, assuming the interquartile range of a certain vibration frequency data segment is calculated to be 8Hz for the first quartile and 12Hz for the third quartile, the upper threshold might be set at 15Hz and the lower threshold at 5Hz. If a sampling point has a value of 16Hz, exceeding the upper threshold, it is considered an outlier and removed, generating a second intermediate dataset with null defects. This method dynamically adapts to changes in equipment operating status, avoiding misjudgments caused by fixed thresholds.

[0038] Linear interpolation is a simple and effective method for addressing missing data in the second intermediate dataset. Suppose that in a time series, the vibration frequency data at a certain time point is missing; the data at the previous time point is 10Hz, and the data at the next time point is 12Hz. Then, linear interpolation can estimate the missing data point to be 11Hz. This method maintains data continuity and generates a complete standardized dataset.

[0039] Specifically, the processing methods in each of the above steps are closely related to the characteristics of the operating parameters of the kitchen waste disposer. Sliding window smoothing effectively reduces noise interference in subsequent analysis, dynamic threshold range setting flexibly addresses different operating states of the equipment, and linear interpolation ensures data integrity. These processing methods work together to ensure that the reliability of the data is improved during the transformation from raw data to a standardized dataset, providing a solid foundation for subsequent equipment status analysis.

[0040] Preferably, in step S2, a short-time Fourier transform is used to perform time-frequency analysis on the tool vibration frequency data in the standardized dataset, the window function is set to a Hanning window, the window length is a preset sampling point, and the time-frequency distribution matrix is ​​obtained.

[0041] For the time-frequency distribution matrix, divide it into multiple frequency bands for energy integration, and calculate the energy proportion of each frequency band as preliminary spectral characteristics;

[0042] From the preliminary spectral features, wavelet decomposition technology is used to extract the core vibration feature frequency bands, and the average value and variance of the core vibration feature frequency bands are calculated to form the spectral features of the vibration signal.

[0043] For time-frequency analysis of blade vibration frequency data, the short-time Fourier transform (SFT) is a tool that decomposes a time-domain signal into two dimensions: time and frequency. Setting the window function to a Hanning window with an assumed window length of 256 sampling points achieves a balance between time and frequency resolution. The time-frequency distribution matrix can intuitively reflect the frequency changes of the vibration signal over different time periods. For example, when a kitchen waste disposer processes hard waste, the frequency may be concentrated in a higher frequency band at a certain time period, while when processing soft waste, it may be concentrated in a lower frequency band. This analysis method helps to capture the dynamic changes in the operating state of the kitchen waste disposer.

[0044] Based on the time-frequency distribution matrix, dividing the frequency range into multiple bands and calculating the energy proportion of each band allows for the further extraction of preliminary spectral features. For example, assuming the frequency range is divided into multiple bands such as 0 to 50 Hz and 50 to 100 Hz, the ratio of the sum of the energy in each band to the total energy is calculated. If the proportion of a certain band is significantly higher than other bands, it may reflect the vibration characteristics of the kitchen waste disposer under specific operating conditions. This method can provide targeted data support for subsequent feature extraction.

[0045] Based on the preliminary spectral characteristics, wavelet decomposition technology is used to extract the core vibration characteristic frequency bands, allowing for a more refined analysis of the signal's local properties. Assuming that wavelet decomposition breaks down the signal into multiple levels, the core vibration characteristic frequency bands related to the key vibration modes of the kitchen waste disposer are extracted, and the mean and variance of these frequency bands are calculated as statistical indicators. These indicators reflect the stability and trend of the vibration signal, providing crucial information for assessing the equipment's condition.

[0046] Optionally, in step S2, a fast Fourier transform is performed on the motor current signal in the motor operating status data of the standardized dataset to analyze multiple harmonic components, calculate the amplitude ratio of each harmonic, and obtain the motor current harmonic characteristics.

[0047] For the current signal in motor operating status data, the Fast Fourier Transform (FFT) can decompose it into multiple harmonic components. Assuming the analysis shows a fundamental amplitude of 10 and a second harmonic amplitude of 2, the ratio of the second harmonic to the fundamental amplitude is 0.2. If this ratio increases significantly over certain time periods, it may reflect changes in motor load or potential operational abnormalities. Extracting this harmonic characteristic helps to understand the working status of the kitchen waste disposer blades from a current perspective.

[0048] Specifically, in step S2, based on the vibration signal spectrum features and the motor current harmonic features, a multidimensional feature set is generated by aligning the timestamps and using a horizontal splicing method. The multidimensional feature set includes the vibration signal spectrum features and the motor current harmonic features.

[0049] Based on the spectral characteristics of vibration signals and the harmonic characteristics of motor current, a multi-dimensional feature set is generated by horizontally stitching together features through timestamp alignment. This method can integrate features from different sources into a unified feature matrix. Assuming that at a certain time point, the statistical indicators of the vibration signal are represented by one set of values, and the current harmonic ratios by another set of values, these are stitched together to form a multi-dimensional vector. This method of constructing feature sets can comprehensively characterize the operating status of the equipment, laying the foundation for subsequent analysis.

[0050] It is worth noting that in step S3, the multidimensional feature set is processed using a standardization method to obtain a standardized feature dataset.

[0051] The standardized feature dataset is input into a pre-trained long short-term memory network model, and time-series analysis is performed on the standardized feature dataset to obtain an evaluation value of the tool wear degree.

[0052] The spectral characteristics of vibration signals reflect the dynamic response of the cutting head during operation, while the harmonic characteristics of the motor current reveal changes in the motor load. Standardization of the multidimensional feature set aims to eliminate the influence of different feature dimensions and ensure data comparison on the same scale. One possible implementation is the mean-standard deviation standardization method, which adjusts the spectral and harmonic characteristic values ​​of the vibration signal to a distribution range with a mean of 0 and a standard deviation of 1. Assuming the average value of each core vibration characteristic frequency band of the cutting head vibration frequency data is between 10 and 50, and the amplitude ratio of each harmonic of the motor current signal ranges from 0.1 to 0.5, after standardization, both are mapped to similar numerical ranges. This method effectively avoids the excessive influence of a single feature's large numerical range on subsequent models.

[0053] When inputting standardized feature datasets into a pre-trained Long Short-Term Memory (LSTM) network model for time-series analysis, the focus is on capturing the dependencies between data over time. LTM models excel at processing time-series data, remembering long-term dependencies and ignoring irrelevant noise. For example, if the vibration signal spectrum of a cutting tool's vibration frequency data shows periodic high-frequency fluctuations over a certain time period, while the current harmonics exhibit a gradually increasing trend, the LTM model can analyze these temporal variations to infer that the cutting tool may be showing signs of wear due to prolonged high-load operation. The model's output evaluation value can be a value between 0 and 1, such as 0.7, indicating a high degree of cutting tool wear.

[0054] In this embodiment, historical operating data and corresponding historical tool wear evaluation values ​​are obtained. The historical operating data includes historical vibration signal spectral characteristics and historical motor current harmonic characteristics. A timestamp alignment method is used to align the evaluation values ​​and the historical operating data in terms of time dimension, resulting in a time-series basic dataset. The time-series basic dataset is input into a Long Short-Term Memory (LSTM) network model for training, resulting in a pre-trained LTM network model. The historical vibration signal spectral characteristics and historical motor current harmonic characteristics are used as inputs, and the historical tool wear evaluation values ​​are used as outputs.

[0055] Preferably, in step S4, the step of matching historical maintenance records according to the multidimensional feature set includes: extracting the vibration signal spectrum features and motor current harmonic features corresponding to the evaluation value of the tool wear degree exceeding the preset threshold from the multidimensional feature set, and generating an abnormal feature dataset;

[0056] Historical records are retrieved from a pre-established database, which includes historical data. A similarity matching method is used to determine historical data that matches the abnormal feature dataset as a subset of historical data. The subset of historical data includes historical vibration signal spectral features and historical motor current harmonic features.

[0057] When retrieving historical feature vectors from a database and performing similarity matching, the Euclidean distance method can be used to calculate the closeness between the current anomalous feature dataset and historical data, and then select the closest historical data as a subset of historical data. If there is a subset of historical data whose feature values ​​deviate from the current anomalous features by less than 5%, it is selected as the matching result. This matching method can quickly filter out relevant historical data, providing a reference for subsequent analysis.

[0058] Optionally, in step S4, the step of calculating the remaining lifetime prediction interval using the sliding window weighted algorithm includes: for the matched historical data subset, segmenting the features in the historical data using a sliding window method to obtain a segmented feature sequence;

[0059] Based on the segmented feature sequence, a weighted method is used to calculate the weighted feature value corresponding to each segmented feature sequence according to the weights of the historical vibration signal spectrum features and historical motor current harmonic features in the historical data subset.

[0060] The weighted feature values ​​and the corresponding timestamp data of the segmented feature sequences are processed using time series analysis methods to obtain feature change trend parameters;

[0061] The historical records also include historical maintenance records. The weighted feature value at the time of tool head replacement in the historical maintenance records is obtained. The time required for the current weighted feature value to reach the weighted feature value at the time of tool head replacement in the historical maintenance records is calculated based on the feature change trend parameter, and used as the remaining life prediction value.

[0062] Sliding window segmentation for historical data subsets can divide the data into segments of 10 seconds each, resulting in multiple segmented feature sequences. This method can capture local variation characteristics of the data over time, laying the foundation for subsequent weight calculations. For example, if the vibration characteristics fluctuate significantly within a certain window, it may reflect that the cutting head was subjected to a high load during that time period.

[0063] When calculating the weights of each feature and generating the weighted eigenvalue, the weights can be assigned based on the contribution of each feature to the wear effect. Assuming the weight of the historical vibration signal spectrum feature is 0.6 and the weight of the historical current harmonic feature is 0.4, the final weighted eigenvalue is 0.52. This weighting method highlights the influence of key features and improves the relevance of the analysis.

[0064] When constructing a trend prediction model, since the segmented feature sequences corresponding to the weighted eigenvalues ​​have unique timestamp data, the weighted eigenvalues ​​and timestamp data can be combined to analyze the changing patterns of the features over time. Assuming that the weighted eigenvalue gradually increased from 0.2 to 0.4 over the past week, the trend parameter obtained through time series analysis is a daily average growth rate of 0.03. This trend parameter can provide important information for subsequent predictions.

[0065] Historical maintenance records matching the historical data corresponding to the weighted feature values ​​are obtained from the historical records. The weighted feature values ​​at the time of tool head replacement are extracted from the historical maintenance records. The difference between the current weighted feature value and the weighted feature value at the time of tool head replacement is calculated. Based on the difference and the current feature change trend parameter, the time from the change of the current weighted feature value to the weighted feature value at the time of tool head replacement is obtained. This time is used as the remaining life prediction value.

[0066] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A method for predicting the lifespan of kitchen waste disposer blades using a deep learning algorithm, characterized in that, The method comprises the following steps: S1: acquiring real-time collected working condition parameters of the kitchen garbage disposer, and generating a standardized data set containing the working condition parameters, the working condition parameters including motor operating state data and cutter head vibration frequency data; S2: performing time-frequency joint analysis on the standardized data set, extracting vibration signal spectrum features and motor current harmonic features, and constructing a multi-dimensional feature set; In the step S2, short-time Fourier transform is used to perform time-frequency analysis on the cutter head vibration frequency data in the standardized data set, a Hanning window is set as the window function, and the window length is a preset sampling point, so as to obtain a time-frequency distribution matrix; For the time-frequency distribution matrix, multiple frequency bands are divided for energy integration, and the energy proportion of each frequency band is calculated as a preliminary spectrum feature; From the preliminary spectrum feature, a wavelet decomposition technique is used to extract a core vibration feature frequency band, the average value and variance of the core vibration feature frequency band are calculated, and a vibration signal spectrum feature is constructed; In the step S2, fast Fourier transform is performed on the motor current signal in the motor operating state data in the standardized data set, a plurality of harmonic components are analyzed, the amplitude ratio of each harmonic is calculated, and a motor current harmonic feature is obtained; In the step S2, based on the vibration signal spectrum feature and the motor current harmonic feature, the timestamps are aligned, and a horizontal splicing method is used to generate a multi-dimensional feature set, the multi-dimensional feature set containing the vibration signal spectrum feature and the motor current harmonic feature; S3: inputting the multi-dimensional feature set containing the vibration signal spectrum feature and the motor current harmonic feature into a pre-trained long short-term memory network model, and outputting an evaluation value of the cutter head wear degree; S4: extracting the vibration signal spectrum feature and the motor current harmonic feature corresponding to the evaluation value of the cutter head wear degree exceeding the preset threshold, and calculating a remaining life prediction value by using a sliding window weighting algorithm according to the multi-dimensional feature set matching historical records.

2. The method of claim 1, wherein the method further comprises: In the step S1, after the working condition parameters are acquired, noise elimination and abnormal value processing are performed on the working condition parameters, and a standardized data set including time-stamped motor operating state data and cutter head vibration frequency data is generated.

3. The method of claim 1, wherein the method further comprises: In the step S3, the multi-dimensional feature set is processed by using a standardization method to obtain a standardized feature data set; The standardized feature data set is input into a pre-trained long short-term memory network model, time series analysis is performed on the standardized feature data set, and an evaluation value of the cutter head wear degree is obtained.

4. The method of claim 1, wherein the method further comprises: In the step S4, the step of matching historical maintenance records according to the multi-dimensional feature set comprises: extracting the vibration signal spectrum feature and the motor current harmonic feature corresponding to the evaluation value of the cutter head wear degree exceeding the preset threshold from the multi-dimensional feature set, and generating an abnormal feature data set; The historical records are retrieved from a pre-established database, the historical records including historical data, a similarity matching method is used to determine historical data matched with the abnormal feature data set as a historical data subset, and the historical data subset includes historical vibration signal spectrum features and historical motor current harmonic features.

5. The method of claim 4, wherein the method further comprises: In the step S4, the step of calculating the residual life prediction interval by using the sliding window weighting algorithm comprises: for the matched historical data subset, segmenting the features in the historical data by using the sliding window method to obtain segmented feature sequences; According to the segmented feature sequences, the weighted method is used to calculate the weighted feature values corresponding to each segmented feature sequence according to the historical vibration signal spectral feature and the historical motor current harmonic feature weight in the historical data subset; The time series analysis method is used to process the time stamp data corresponding to the weighted feature values and the corresponding segmented feature sequences to obtain feature change trend parameters; The historical record further comprises a historical maintenance record, the weighted feature value when the tool bit is replaced in the historical maintenance record is obtained, and the time required for the current weighted feature value to the weighted feature value when the tool bit is replaced in the historical maintenance record is calculated according to the feature change trend parameters, as a residual life prediction value.

Images