A method for determining the grinding wheel wear state of a multi-station scalpel grinding machine

Abstract

The present application relates to the technical field of grinding wheel wear prediction of knife grinder, and discloses a kind of grinding wheel wear state determination method of multi-station scalpel blade grinding machine, collects acoustic emission source signal data, grinding power and abrasive grain height of grinding wheel, carries out pre-processing and obtains data set by feature extraction;Acoustic emission source signal and grinding power are used as the input variable of random forest model, and the abrasive grain height of the grinding wheel is used as the output variable, to establish RF detection model;The CPA algorithm is improved, and the optimal RF detection model is obtained by optimizing the hyperparameters of the RF detection model using ICPA;The optimal RF detection model is used to detect the grinding wheel wear state with the current collected acoustic emission source signal and grinding power as input, to obtain the abrasive grain height of the grinding wheel under the current state, and further obtain the wear state of the current grinding wheel.Compared with the prior art, the present application provides a kind of grinding wheel wear state determination method, so that the control system more simply obtains the feed amount of grinding wheel, thereby compensating the grinding wheel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of grinding wheel wear prediction of knife grinder, and particularly relates to a grinding wheel wear state determination method of a multi-station surgical blade grinding machine based on ICPA optimized RF model. BACKGROUND

[0002] Surgical blades are commonly used disposable medical machines in hospitals and are indispensable important surgical tools in surgical operations, with a global market demand of 2 billion pieces. Traditional blade sharpening is single-station manual grinding, which is low in efficiency and extremely dangerous. The emergence of surgical blade grinding machines increases the efficiency of surgical blade production, but also has technical deficiencies. For the multi-station surgical blade grinding machine, the grinding wheel compensation accuracy cannot reach the expected ideal effect, and when the grinding wheel grinding state is abnormal, it is easy to cause the blade width to be too large or too small, affecting the normal use of the surgical blade. Therefore, through deep learning, the grinding wheel wear state can be accurately predicted, so that the grinding wheel can be compensated accordingly, and the production efficiency of the surgical blade can be improved.

[0003] Random forest (Random forest, RF) is a classifier containing multiple decision trees, which belongs to a kind of machine learning, and has made a major breakthrough in classification and regression operations. Random forest obtains the output class of decision tree or the average prediction value of all single decision trees by constructing a large number of decision trees during training and using a voting mechanism. Random forest randomly samples data and features with replacement, constructs multiple decision trees, and integrates the final result according to the combination strategy. Random forest can handle a large number of input variables and has good noise resistance, simple operation and other advantages. However, random forest also has overfitting and long training time, so a more efficient optimization algorithm is needed to improve the accuracy of random forest and meet the model prediction requirements.

[0004] The Carnivorous Plant Algorithm (CPA), proposed by Ong KokMeng of Malaysia in 2020, is inspired by how carnivorous plants adapt to harsh environments (such as preying on insects and pollinating for reproduction). It's a novel mathematical model that simulates the attraction, trapping, digestion, and reproduction strategies of carnivorous plants. Existing CPA algorithms suffer from slow convergence, susceptibility to local optima, and low result quality. An excellent metaheuristic algorithm should possess the following characteristics: first, fully utilize the positional information of individuals within the population; second, leverage the guidance of superior or optimal individuals; third, have strong focus on different stages of exploration and development; and finally, maintain population diversity. To address the shortcomings of the CPA algorithm, this paper introduces Latin hypercube sampling initialization, adaptive adjustment of uptake and growth rates, and Cauchy mutation operations to optimize the CPA algorithm. The optimized CPA algorithm is then used to further optimize some hyperparameters in a random forest model, reconstructing a new random forest model to accurately predict the wear degree of grinding wheels. Summary of the Invention

[0005] Purpose of the invention: To address the problems existing in the prior art, this invention proposes a method for determining the wear state of the grinding wheel in a multi-station surgical blade grinding machine. This method solves the technical problems of difficulty in determining parameters in existing RF regression models, as well as insufficient search and easy getting trapped in local optima in CPA algorithms. It provides a method for determining the wear state of the grinding wheel, and further enables the control system to obtain the feed amount to the grinding wheel, thereby compensating for the grinding wheel wear.

[0006] Technical solution: This invention provides a method for determining the wear state of the grinding wheel in a multi-station surgical blade grinding machine, comprising the following steps:

[0007] Step 1: Acoustic emission sensors are used to collect data on the acoustic emission sources during the grinding process. Power sensors are used to collect the voltage and current of the spindle motor during the grinding process and calculate the grinding power. Laser displacement sensors are used to measure the abrasive grain height of the grinding wheel and use it as an evaluation index of grinding wheel wear. The emission waveform of the acoustic emission source is preprocessed and a dataset is obtained through feature extraction.

[0008] Step 2: Preprocess the acoustic emission source signal, grinding power, and the corresponding abrasive grain height of the grinding wheel. Select training set samples and test set samples. Use the acoustic emission source signal and grinding power as input variables of the random forest model, and the abrasive grain height of the grinding wheel as output variable of the random forest model to establish an RF detection model.

[0009] Step 3: The CPA algorithm is improved, and Latin hypercube sampling is used to initialize the CPA; the absorption rate and growth rate are adaptively changed; the position of the optimal carnivorous plant is introduced into the Cauchy mutation operation to obtain ICPA;

[0010] Step 4: The hyperparameters of the RF detection model are optimized using ICPA to obtain the optimal hyperparameters, and further obtain the optimal RF detection model;

[0011] Step 5: Using the optimal RF detection model, the current collected acoustic emission source signal and grinding power as input to detect the grinding wheel wear state, obtain the abrasive particle height of the grinding wheel under the current state, and further obtain the wear state of the current grinding wheel.

[0012] Further, the specific operation of preprocessing and feature extraction of the acoustic emission source waveform in step 1 to obtain the data set is:

[0013] Step 1.1: The wavelet coefficients are processed by a soft threshold function, and the threshold calculation formula is , wherein is the standard deviation of the noise signal, is the length of the signal; the wavelet coefficients containing the noise signal are compared with the threshold , when the absolute value of the wavelet coefficient is greater than the given threshold, the wavelet coefficient is subtracted from the threshold; when it is less than the threshold, the wavelet coefficient is zero;

[0014] Step 1.2: For threshold selection, four typical threshold thresholds, unbiased risk estimation threshold (rigrsure), fixed threshold (sqtwolog), heuristic threshold (heursure), and maximum minimum threshold (minimaxi), are used to denoise the signal in the grinding process. Finally, the method that best removes noise without distorting the signal is selected;

[0015] Step 1.3: According to the decomposition coefficients and quantization coefficients of the wavelet packet, the wavelet packet is reconstructed to achieve the purpose of eliminating noise;

[0016] Step 1.4: Feature extraction is performed on the denoised waveform, and the extracted data is used as an input variable of the model.

[0017] Further, the grinding particle height of the grinding wheel is measured using a laser displacement sensor in step 1, specifically: when a specified material is worn, the laser displacement sensor measures the abrasive particle height of multiple regions in a certain direction , and calculates the average value to finally obtain the grinding particle height of the grinding wheel, and the specific calculation formula is:

[0018]

[0019] wherein, is the number of data points in a certain area; is the position data of abrasive grains in a certain area measured by a laser displacement sensor; is the average value of all data in the area.

[0020] Further, the specific process of establishing the RF detection model in step 2 is as follows:

[0021] Step 2.1: Randomly sort the obtained input variables and corresponding output variables, and use 70% of the sorted variables as training set samples and the last 30% as test set samples;

[0022] Step 2.2: Before training the model, the input variables and output variables need to be normalized;

[0023] Step 2.3: Use Bootstrap sampling to randomly extract a sample with replacement each time, construct a new sample set S, S

[0024] Step 2.4: Prune each decision tree;

[0025] Step 2.5: Combine all decision trees into a random forest, and use the average value of the sum of the detection results of each tree as the final result of the random forest.

[0026] Further, the improvement of the CPA algorithm in step 3 includes:

[0027] Step 3.1: Determine the population number of the CPA algorithm as nPop and the dimension D, where nPop = nCP + nPrey, nCP is the number of carnivorous plants, and nPrey is the number of prey;

[0028] Step 3.1.1: Generate nPop equal parts in [0, 1], where the upper boundary of each part is Ub and the lower boundary is Lb, and uniformly distribute a random number in the range:

[0029]

[0030]

[0031] Step 3.1.2: Shuffle the order of the nPop random numbers, and use these n numbers as the probability of each random sample, and generate the corresponding values of the population according to the inverse function of the probability distribution function;

[0032] Step 3.2: Optimize the growth rate and absorption rate, and the calculation formulas for the adaptive changes of the attraction rate from small to large and the growth rate from large to small are as follows:

[0033] Step 3.2.1: For the absorption rate, at the beginning of iteration, the absorption rate should be made to be minimum, and with the iteration, the absorption rate should gradually increase, and after using the adaptive function optimization, there is:

[0034] wherein, is the maximum absorption rate, is the maximum number of iterations, is the current number of iterations;

[0035] Step 3.2.2: For the growth rate, there is an initial growth rate maximum, and gradually decreases with the number of iterations, and after using the adaptive function optimization, there is:

[0036] wherein, is the minimum growth rate;

[0037] Step 3.3: Perform Cauchy mutation operation on the optimal carnivorous plant to increase population diversity and increase search space;

[0038] Step 3.3.1: The standard Cauchy distribution function is as follows, and the Cauchy formula for the optimal carnivorous plant population is as follows:

[0039]

[0040]

[0041] wherein, is the generation optimal carnivorous plant population;

[0042] Step 3.3.2: Use the greedy selection mechanism to compare the fitness values of the mutated optimal carnivorous plant population and the optimal carnivorous plant population before mutation, if the mutated one is better than the one before mutation, replace the population, otherwise keep it unchanged:

[0043] .

[0044] Further, the step 4 utilizes ICPA to optimize the hyperparameters of the RF detection model, specifically:

[0045] Step 4.1: Use the ICPA algorithm to optimize the number of decision trees and the maximum depth max_depth and the maximum leaf node number max_leaf_nodes of the random forest;

[0046] ​Step 4.2: After the initialization of the ICPA algorithm, the initialized numerical value is used to build a random forest detection model, the model is trained with the training set, and the model after training is tested with the test set to obtain the prediction result, and the mean square error of the predicted value and the actual value is calculated:

[0047]

[0048] Wherein, n is the number of prediction samples, is the actual grinding wheel wear at time i, is the predicted grinding wheel wear at time i;

[0049] Step 4.3: The root mean square error of the random forest detection model prediction value and the corresponding actual value is used as the fitness value function of the ICPA algorithm, and the fitness value function is used as the basis for grouping, growing and breeding of the ICPA algorithm.

[0050] Beneficial effects:

[0051] The present application uses grinding power and acoustic emission source signal as model input, uses abrasive grain height of the grinding wheel as output variable of the random forest model, and the grinding wheel can indirectly understand the grinding wheel wear condition through the physical change amount in the grinding process. The abrasive grain height can directly understand the grinding wheel wear condition. Multiple signal input can solve the problems of small sample data amount, low training and prediction precision and the like in reaction of the grinding wheel wear state by single signal. And using the abrasive grain height as the output variable can more directly understand the grinding wheel wear condition, and provide input data support for subsequent grinding wheel compensation. The present application improves the CPA algorithm, and solves the problems of slow iteration speed and easy to fall into local optimum in the convergence process of the CPA algorithm. The improved CPA algorithm is used to optimize the hyperparameters of the RF detection model, and solves the problems of overfitting and long training time in the training process of the model. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a flow chart of the grinding wheel wear detection method of the multi-station surgical blade grinding machine of the present application;

[0053] Figure 2 is a data collection flow chart of the present application;

[0054] Figure 3 is a flow chart of the grinding wheel wear detection and nonlinear compensation of the grinding wheel wear according to the grinding wheel feed amount of the multi-station surgical blade grinding machine of the present application. DETAILED DESCRIPTION

[0055] The present application will be further described below in conjunction with the drawings. The following examples are only used to more clearly illustrate the technical solutions of the present application, and cannot be used to limit the protection scope of the present application.

[0056] The application discloses a method for determining the wear of a grinding wheel of a multi-station surgical blade grinding machine, and refers to Figure 1 , comprising the following steps.

[0057] Step 1: During the operation of the multi-station surgical blade grinding machine, an acoustic emission sensor is used to collect acoustic emission signals during the grinding process of the grinding wheel, and the collected signals are converted into electrical signals. After amplification, filtering and signal processing, the acoustic emission waveform is recorded and displayed. Finally, through signal analysis and processing, characteristic parameters are extracted to evaluate the characteristics of the signal source. A power sensor is used to collect the voltage and current of the spindle motor during the grinding process, and the grinding power input variable is calculated. The grinding wheel wear evaluation index is directly represented by the height of the abrasive particles on the surface of the grinding wheel, and a laser sensor is used to measure and calculate the height of the abrasive particles on the surface of the grinding wheel.

[0058] Step 1.1: In order to collect acoustic emission signals that can truly reflect the wear state of the grinding wheel, noise reduction processing is required for the acoustic emission signals. Wavelet processing is selected for noise reduction, and orthogonal decomposition is performed on the signal at low and high frequencies to obtain the frequency components in any frequency band.

[0059] Step 1.1.1: Soft threshold function processing is performed on the wavelet coefficients. The threshold calculation formula is:

[0060] (1)

[0061] wherein is the standard deviation of the noise signal, is the length of the signal. The wavelet coefficients containing the noise signal are compared with the threshold . When the absolute value of the wavelet coefficient is greater than the given threshold, the wavelet coefficient is subtracted from the threshold; when it is less than the threshold, the wavelet coefficient is set to zero.

[0062] Step 1.1.2: For threshold selection, four typical threshold values, unbiased risk estimation threshold (rigrsure), fixed threshold (sqtwolog), heuristic threshold (heursure) and maximum minimum threshold (minimaxi), are used to denoise the signals during the grinding process. Finally, the method with the best noise reduction effect is selected on the premise that the signal is not easily distorted.

[0063] Step 1.1.3: The wavelet packet is reconstructed according to the decomposition coefficients and quantization coefficients of the wavelet packet, so as to achieve the purpose of eliminating noise.

[0064] Step 1.1.4: Feature extraction is performed on the denoised waveform, and the extracted data is used as the input set of the model.

[0065] Step 1.2: Use a power sensor to collect the voltage and current of the spindle motor during the grinding process, and calculate the grinding power input variable.

[0066] Step 1.3: Use a laser displacement sensor to measure the abrasive grain height of the grinding wheel, which will be used as an indicator of the grinding wheel's wear. Under the action of pressure and friction, the edges of the abrasive grains on the grinding wheel gradually become blunt or break, and the abrasive grain height will gradually decrease.

[0067] Step 1.3.1: The formula for calculating the abrasive grain height of the grinding wheel is:

[0068] (2)

[0069] in The number of data points within a certain area; The laser displacement sensor measures the position data of abrasive grains in a certain area of ​​the grinding wheel; This is the average value of all data within the region.

[0070] Step 1.3.2: After the specified material has been worn away, use a laser displacement sensor to measure the abrasive grain height in multiple areas along a certain direction. The average value was calculated, and the final abrasive grain height of the grinding wheel was obtained.

[0071] Step 2: Use the acoustic emission source dataset measured by the acoustic emission sensor and the grinding power input variable as input variables of the random forest model, and the grinding wheel abrasive height measured by the laser displacement sensor as output variables of the random forest model to construct the RF model.

[0072] Step 2.1: Randomly sort the input variables and corresponding output variables. 70% of the sorted variables are used as training set samples, and the remaining 30% are used as test set samples.

[0073] Step 2.2: Before training the RF model, the input and output variables need to be normalized to prevent a single feature from dominating or overfitting during training. The normalization method is as follows:

[0074] (3)

[0075] in, These are the normalized data values. For and These are the maximum and minimum values ​​in the original data, respectively, and the normalized range is [0,1].

[0076] Step 2.3: Bootstrap sampling is used to sample one sample at a time with replacement to construct a new sample set S (S < N) as a new training set, and a decision tree is constructed using the new training set.

[0077] Step 2.4: Pruning is performed on each decision tree.

[0078] Step 2.5: All decision trees are combined into a random forest, and the average of the sum of detection results of each tree is taken as the final detection result of the random forest.

[0079] Step 3: Improvement of CPA algorithm, using Latin hypercube sampling to initialize CPA; adaptive change of absorption rate and growth rate; introduction of Cauchy mutation operation to the position of optimal carnivorous plant to obtain ICPA.

[0080] Step 3.1: Use Latin hypercube sampling to initialize the population of CPA algorithm. Determine the population size of CPA algorithm as nPop and dimension D, where nPop = nCP + nPrey, nCP is the number of carnivorous plants, and nPrey is the number of prey.

[0081] Step 3.1.1: Generate nPop equal parts in [0, 1], where the upper boundary of each part is Ub and the lower boundary is Lb. The upper and lower limits are calculated by formulas (4) and (5), and a number is uniformly randomly generated in the range:

[0082] (4)

[0083] (5)

[0084] Step 3.1.2: Shuffle the order of the nPop random numbers, and use the n numbers as the probability of each random sample. According to the inverse function of the probability distribution function, generate the corresponding values of the population.

[0085] Step 3.2: For the growth process of CPA algorithm, the absorption rate and growth rate are both constants, which do not conform to the current life. Therefore, the growth rate and absorption rate are optimized. The calculation formulas for the adaptive change of attraction rate from small to large and the adaptive change of growth rate from large to small are given.

[0086] Step 3.2.1: For the absorption rate, at the beginning of iteration, the absorption rate should be minimum, and with iteration, the absorption rate should gradually increase. After optimization using the adaptive function, formula (6) is obtained:

[0087] (6)

[0088] where maximizing the absorption rate, maximizing the number of iterations, is the current number of iterations.

[0089] Step 3.2.2: For the growth rate, there is an initial growth rate maximizing the number of iterations, and gradually decreasing with the number of iterations, the formula (7) after optimization using the adaptive function has:

[0090] (7)

[0091] where is the minimum growth rate.

[0092] Step 3.3: When solving the optimization problem using the CPA algorithm, the carnivorous plants tend to gather around some extreme points in the later iterations, and are prone to falling into local optima. In order to reduce the probability of falling into local stagnation, Cauchy mutation operation is performed on the optimal carnivorous plants to increase population diversity and search space.

[0093] Step 3.3.1: The standard Cauchy distribution function formula (8) is as follows, and the Cauchy formula (9) is constructed for the optimal carnivorous plant population as follows:

[0094] (8) (9)

[0095] where is the optimal carnivorous plant population of the generation.

[0096] Step 3.3.2: Using the greedy selection mechanism, the fitness values of the mutated optimal carnivorous plant population and the unmutated optimal carnivorous plant population are compared. If the mutated one is better than the unmutated one, the population is replaced, otherwise it remains unchanged.

[0097] (10)

[0098] Step 4: The optimized carnivorous plant algorithm ICPA is used to optimize some hyperparameters in the random forest model.

[0099] Step 4.1: The optimized carnivorous plant algorithm ICPA is used to optimize the number of decision trees in the random forest and the maximum depth max_depth and the maximum leaf node number max_leaf_nodes. Through the above optimization, the model can be prevented from overfitting during training, and the fitting and prediction time can be accelerated.

[0100] Step 4.2: After the initialization of the improved carnivorous plant algorithm ICPA, the initialized values are used to build a random forest model, the model is trained with the training set, and the trained model is tested with the test set to obtain the prediction results, and the mean square error of the predicted value and the actual value is calculated:

[0101] (11)

[0102] where n is the number of prediction samples, is the actual grinding wheel wear at time i, is the predicted grinding wheel wear at time i.

[0103] Step 4.3: The root mean square error of the RF model prediction value and the corresponding actual value is used as the fitness value function of the ICPA algorithm, which is used as the basis for grouping, growing and breeding of the ICPA algorithm.

[0104] (12)

[0105] Step 5: The optimal hyperparameters of the random forest model are obtained, and the test sample data is input into the RF model to obtain the prediction results. The optimal RF detection model is used to detect the grinding wheel wear state with the current collected acoustic emission source signal and grinding power as input, to obtain the abrasive particle height of the grinding wheel under the current state, and further to obtain the wear state of the grinding wheel.

[0106] Finally, the data is uploaded to the database through the cloud through the 5G router for model training, and the predicted grinding wheel wear value is input into the PLC control system to calculate the grinding wheel feed amount, and finally the terminal stepping motor of the PLC control driver is used to perform nonlinear compensation on the grinding wheel wear.

[0107] The above embodiments are only for illustrating the technical concept and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the content of the present application and implement it, and cannot limit the protection scope of the present application. Any equivalent transformation or modification made in accordance with the spirit and essence of the present application should be covered within the protection scope of the present application.

Claims

1. A method of determining the state of wear of a grinding wheel of a multi-station surgical blade grinding machine, characterized by, It comprises the following steps: Step 1: Collecting data of acoustic emission source in the grinding process of the grinding wheel by using acoustic emission sensor, collecting voltage and current of the spindle motor in the grinding process by using power sensor, calculating the grinding power, measuring the height of abrasive particles of the grinding wheel by using laser displacement sensor, and taking it as the evaluation index of the grinding wheel wear, pre-processing the emission waveform of the acoustic emission source, and obtaining the data set through feature extraction; Step 2: Preprocessing the acoustic emission source signal, the grinding power and the height of abrasive particles of the corresponding grinding wheel, selecting the training set sample and the test set sample, taking the acoustic emission source signal and the grinding power as the input variables of the random forest model, and taking the height of abrasive particles of the grinding wheel as the output variable of the random forest model, and establishing the RF detection model; Step 3: Improving the CPA algorithm, initializing the CPA by using Latin hypercube sampling, and introducing Cauchy mutation operation to the position of the optimal carnivorous plant to obtain ICPA; Step 4: Using ICPA to optimize the hyperparameters of the RF detection model to obtain the optimal hyperparameters, and further obtaining the optimal RF detection model; using ICPA algorithm to optimize the number of decision trees and the maximum depth max_depth and the maximum leaf node number max_leaf_nodes of the tree of the random forest; Step 5: Using the optimal RF detection model to detect the grinding wheel wear state by taking the currently collected acoustic emission source signal and grinding power as the input, obtaining the height of abrasive particles of the grinding wheel under the current state, and further obtaining the wear state of the current grinding wheel.

2. The method of claim 1, wherein the method further comprises: The specific operation of preprocessing the emission waveform of the acoustic emission source in step 1 to obtain the data set through feature extraction is as follows: Step 1.1: The soft threshold function is applied to the wavelet coefficients, and the threshold is calculated as wherein is the standard deviation of the noise signal, is the length of the signal; the wavelet coefficients containing the noise signal are compared with the threshold , and when the absolute value of the wavelet coefficient is greater than the given threshold, the wavelet coefficient is reduced by the threshold; when it is less than the threshold, the wavelet coefficient is set to zero; Step 1.2: For threshold selection, four typical threshold values of unbiased risk estimation threshold (rigrsure), fixed threshold (sqtwolog), heuristic threshold (heursure) and minimax threshold (minimaxi) are used to denoise the signal in the grinding process, and finally the method with the best denoising effect is selected under the premise that the signal is not easy to distort; Step 1.3: Reconstructing the wavelet packet according to the decomposition coefficient and the quantization coefficient of the wavelet packet to eliminate noise; Step 1.4: Extracting features from the denoised waveform, and taking the extracted data as an input variable of the model.

3. The method of claim 1, wherein the method further comprises: determining a state of the grinding wheel based on the measured value of the grinding wheel. The abrasive grain height of the grinding wheel is measured using the laser displacement sensor in step 1, specifically, after a prescribed amount of material is worn away, the abrasive grain height of a plurality of regions is measured in a certain direction using the laser displacement sensor and the average value is calculated to obtain the abrasive grain height of the grinding wheel, specifically, the calculation formula is: ​ ； wherein, is the number of data points in a region; is the position data of abrasive grains in a region measured by a laser displacement sensor; is the average value of all data in the region.

4. The method of claim 1, wherein the method further comprises: The specific process of establishing the RF detection model in step 2 is as follows: Step 2.1: Randomly sorting the obtained input variables and corresponding output variables, and taking 70% of the sorted variables as the training set sample and the last 30% as the test set sample; Step 2.2: Before training the model, the input variables and the output variables need to be normalized; Step 2.3: Using Bootstrap sampling, a new sample set S is constructed by randomly sampling a sample with replacement each time, and S is used as the new training set to construct a decision tree; Step 2.4: Pruning each decision tree; Step 2.5: Combining all decision trees into a random forest, and taking the average value of the sum of the detection results of each tree as the final result of the random forest.

5. The method of claim 1, wherein, The improvement of the CPA algorithm in step 3 specifically includes: Step 3.1: Determine the population number of the CPA algorithm as nPop and the dimension D, wherein nPop = nCP + nPrey, nCP is the number of carnivorous plants, and nPrey is the number of prey; Step 3.1.1: Generate nPop equal parts in [0, 1], wherein the upper boundary of each part is Ub, and the lower boundary is Lb, and a number is randomly generated in the range: ； ； Step 3.1.2: Then, the order of the nPop random numbers is shuffled, and the n numbers are used as the probability of each random sample, and the corresponding values of the population are generated according to the inverse function of the probability distribution function; Step 3.2: Optimize the growth rate and absorption rate, and the calculation formula of the adaptive change of the attraction rate from small to large and the adaptive change of the growth rate from large to small: Step 3.2.1 : For Absorptivity, at the beginning of the iteration, the absorptivity should be made minimum and as the iteration proceeds, the absorptivity should be gradually increased, using an adaptive function optimization gives: 0.5 + 0.5 * (1 - exp(-0.1 * (1 - Absorptivity)) wherein, is the maximum absorption rate, is the maximum number of iterations, is the current number of iterations; Step 3.2.2: For the growth rate, there is an initial growth rate maximum, and decreases gradually with the number of iterations, after optimization using an adaptive function: ; wherein, is the minimum growth rate; Step 3.3: Perform Cauchy mutation operation on the optimal carnivorous plant to increase population diversity and search space; Step 3.3.1: The standard Cauchy distribution function is as follows, and the Cauchy formula for the optimal carnivorous plant population is as follows: ； ； wherein is optimal carnivorous plant population of the generation; Step 3.3.2: Use the greedy selection mechanism to compare the fitness values of the mutated optimal carnivorous plant population and the optimal carnivorous plant population before mutation. If the mutated one is better than the one before mutation, replace the population, otherwise keep it unchanged: 。 6. The method of determining the state of wear of the grinding wheel of a multi-station surgical blade grinder according to claim 5, wherein, The super parameter of the RF detection model optimized by ICPA in step 4 is specifically: Step 4.1: After the initialization of the ICPA algorithm, the initialized values are used to build a random forest detection model, the model is trained with the training set, and the prediction result is obtained by testing the trained model with the test set. Calculate the mean square error of the predicted value and the actual value: ； wherein n is the number of prediction samples, is the actual grinding wheel wear amount at time i, is the predicted grinding wheel wear amount at time i; Step 4.2: The root mean square error of the random forest detection model prediction value and the corresponding actual value is used as the fitness value function of the ICPA algorithm, which is used as the basis for grouping, growing, and breeding of the ICPA algorithm.

Images