A tool wear state recognition method and system, an electronic device, and a storage medium

Abstract

The application provides a tool wear state recognition method, system, electronic equipment and storage medium, the method comprises the following steps: collecting tool wear vibration signals; inputting the tool wear vibration signals into a preset convolutional neural network, performing feature extraction and one-dimensional dimension reduction, and obtaining vibration signal features; based on a preset local linear embedding rule, performing two-dimensional dimension reduction on the vibration signal features, and obtaining an observation sequence; inputting the observation sequence into a pre-trained hidden Markov model, performing tool wear state recognition and tool wear state prediction, and obtaining recognition results and prediction results; and feeding back the recognition results and prediction results to an associated terminal device. The tool wear state recognition method provided by the application can automatically realize feature extraction of tool wear vibration signals, has high accuracy, strong real-time performance, and high tool wear state recognition accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal recognition, and in particular to a tool wear state recognition method and system, an electronic device, and a storage medium. BACKGROUND

[0002] Tool wear is a common fault in numerical control machining. The machining problems caused by tool wear will directly affect the machining quality, production efficiency and machining cost of the product. In serious cases, it may also cause damage to the machine tool itself and result in significant economic losses. Traditional tool wear state monitoring and detection methods mainly rely on offline measurement or skilled workers' judgment based on personal experience. On the one hand, the monitoring efficiency is low, and on the other hand, it is difficult to obtain accurate and scientific conclusions based on manual experience. Since the twentieth century, science and technology have developed rapidly. The monitoring level of numerical control machining has also improved with the increasing number of sensors and communication devices equipped in machine tools and more advanced and intelligent monitoring methods, achieving more stable and efficient monitoring.

[0003] Current tool wear state recognition methods mainly manually extract features from collected signals and process the extracted features based on simulation models or classifiers to evaluate the tool wear state. However, this manual signal feature extraction method has poor real-time performance and low accuracy, which cannot be well applied to online monitoring of tool wear. Moreover, the feature processing and recognition based on simulation models and classifiers have low accuracy and cannot accurately predict the tool wear state. SUMMARY

[0004] The present application provides a tool wear state recognition method, system, electronic device and storage medium to solve the problem of poor real-time performance and low accuracy caused by manual signal feature extraction in the tool wear state recognition process in the prior art, and the problem that existing tool wear state recognition methods cannot accurately predict the tool wear state.

[0005] The present application provides a tool wear state recognition method, comprising:

[0006] Collecting tool wear vibration signals;

[0007] Inputting the tool wear vibration signals into a preset convolutional neural network for feature extraction and first dimension reduction to obtain vibration signal features;

[0008] Based on a preset local linear embedding rule, the vibration signal features are subjected to second dimension reduction to obtain observation sequences;

[0009] The observation sequence is input into a pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction, and to obtain a recognition result and a prediction result.

[0010] The recognition result and the prediction result are fed back to an associated terminal device.

[0011] Optionally, the tool wear vibration signal is input into a preset convolutional neural network to perform feature extraction and one-dimensional dimension reduction, and the step of obtaining the vibration signal feature includes:

[0012] The tool wear vibration signal is input into a convolutional layer of the convolutional neural network to perform feature extraction and obtain a target feature;

[0013] The target feature is input into a pooling layer of the convolutional neural network to perform one-dimensional dimension reduction and obtain the vibration signal feature.

[0014] Optionally, based on a preset local linear embedding rule, the vibration signal feature is subjected to two-dimensional dimension reduction to obtain an observation sequence, and the step of obtaining the observation sequence includes:

[0015] Based on a preset neighbor point discrimination sub-rule, an initial neighbor point of each sample point in the vibration signal feature is determined, and each sample point corresponds to one or more initial neighbor points;

[0016] At least one target neighbor point is obtained by sorting and screening the initial neighbor points according to the Euclidean distance between the sample points and the corresponding initial neighbor points;

[0017] A local reconstruction weight matrix of the sample point corresponding to the target neighbor point is obtained by fitting a linear relationship of the target neighbor point;

[0018] Based on the local reconstruction weight matrix of each sample point and its target neighbor point, the observation sequence after two-dimensional dimension reduction is obtained.

[0019] Optionally, the step of obtaining the hidden Markov model includes:

[0020] An observation sequence sample set is obtained, and the observation sequence sample set includes one or more observation sequence samples;

[0021] Parameter initialization is performed on a preset original model, and the number of hidden states is set, and the hidden states include initial wear, intermediate wear, and severe wear;

[0022] inputting the observation sequence sample into the initialized original model, using a preset forward-backward algorithm to obtain a first intermediate quantity and a second intermediate quantity, the first intermediate quantity being obtained based on a forward probability and a backward probability of each hidden state of the observation sequence sample, and the second intermediate quantity being obtained based on the forward probability, the backward probability and a state transition probability of each hidden state of the observation sequence sample;

[0023] updating and iterating model parameters of the original model based on the first intermediate quantity and the second intermediate quantity to obtain a trained hidden Markov model.

[0024] Optionally, the step of inputting the observation sequence into the pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction to obtain a recognition result and a prediction result comprises:

[0025] inputting the observation sequence into the hidden Markov model to perform initial forward probability calculation and / or initial backward probability calculation to obtain initial forward probability and / or initial backward probability of the observation sequence;

[0026] forwardly recursively calculating the initial forward probability according to time sequence to obtain target forward probability, and / or backwardly recursively calculating the initial backward probability according to time sequence to obtain target backward probability;

[0027] obtaining the recognition result based on the target forward probability and / or the target backward probability.

[0028] Optionally, the step of inputting the observation sequence into the pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction to obtain a recognition result and a prediction result comprises:

[0029] outputting the observation sequence from the hidden Markov model to obtain initial local state data;

[0030] performing local state dynamic programming based on the initial local state data to obtain predicted local states at multiple time points;

[0031] obtaining target hidden state sequence probability and target hidden state sequence at a target time point based on the predicted local states;

[0032] performing backtracking based on the predicted local states and the target hidden state sequence to obtain a final hidden state sequence; taking the final hidden state sequence as the prediction result.

[0033] Optionally, the step of feeding back the recognition result and the prediction result to an associated terminal device comprises:

[0034] issuing an alarm or generating warning information based on the recognition result and the prediction result.

[0035] feedback the warning information to the associated terminal device.

[0036] The application further provides a tool wear state recognition system, comprising:

[0037] a collection module, configured to collect tool wear vibration signals;

[0038] a convolution module, configured to input the tool wear vibration signals into a preset convolutional neural network to perform feature extraction and one-dimensional dimension reduction, and obtain vibration signal features;

[0039] a dimension reduction module, configured to perform two-dimensional dimension reduction on the vibration signal features based on a preset local linear embedding rule, and obtain observation sequences;

[0040] a recognition and prediction module, configured to input the observation sequences into a pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction, and obtain recognition results and prediction results;

[0041] a communication module, configured to feedback the recognition results and the prediction results to an associated terminal device.

[0042] The application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the tool wear state recognition method according to any one of the above when executing the program.

[0043] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the tool wear state recognition method according to any one of the above.

[0044] The tool wear state recognition method, system, electronic device, and storage medium provided by the application can automatically implement feature extraction of tool wear vibration signals, have high accuracy, and have high real-time performance, and the tool wear state recognition accuracy of the method is high, and the tool wear state can be accurately predicted. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings described below are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0046] Figure 1 is a flowchart of the tool wear state recognition method provided by the present application;

[0047] Figure 2 is a flowchart of the tool wear state recognition method provided by the present application;

[0048] Figure 3 is a flowchart of the tool wear state recognition method provided by the present application;

[0049] Figure 4 is a flowchart of the tool wear state recognition method provided by the present application;

[0050] Figure 5 is a flowchart of the tool wear state recognition method provided by the present application;

[0051] Figure 6 is a flowchart of the tool wear state recognition method provided by the present application;

[0052] Figure 7 is a flowchart of the tool wear state recognition method provided by the present application;

[0053] Figure 8 is a structural diagram of the tool wear state recognition system provided by the present application;

[0054] Figure 9 is a structural diagram of the electronic device provided by the present application. DETAILED DESCRIPTION

[0055] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0056] The following will be described in the manner of embodiments in combination with the drawings in the present application. Figures 1-9The application provides a tool wear state recognition method, a system, an electronic device and a storage medium.

[0057] Please refer to Figure 1 The tool wear state recognition method comprises the following steps.

[0058] S101: Collect a tool wear vibration signal.

[0059] Specifically, the tool wear vibration signal is a vibration signal of tool wear of a numerical control machine tool. The tool wear vibration signal can be collected by installing a vibration sensor on the numerical control machine tool. The sampling rate can be set according to actual conditions, such as 6000 Hz (hertz), and will not be described here. By collecting the tool wear vibration signal of the numerical control machine tool, subsequent recognition and processing based on the collected tool wear vibration signal can be facilitated, the tool wear state can be determined, and real-time monitoring of the tool wear state can be realized.

[0060] S102: Input the tool wear vibration signal into a preset convolutional neural network to perform feature extraction and one-dimensional dimension reduction to obtain a vibration signal feature.

[0061] It should be noted that by inputting the tool wear vibration signal into the preset convolutional neural network to perform feature extraction, automatic extraction of the tool wear vibration signal feature can be better realized, the accuracy is higher, the influence of data noise is effectively reduced, the comprehensiveness of feature extraction is stronger, and the problems of lower accuracy and poorer real-time performance caused by manual or mechanical extraction are avoided. Moreover, by using the convolutional neural network to perform one-dimensional dimension reduction on the target feature obtained through feature extraction, the dimension of the target feature can be reduced to a certain extent, and the complexity of data processing is further reduced.

[0062] S103: Based on a preset local linear embedding rule, perform two-dimensional dimension reduction on the vibration signal feature to obtain an observation sequence.

[0063] Specifically, the local linear embedding rule is that, based on the near neighbor points of each sample point in the vibration signal feature, local linear embedding is performed, that is, any sample point can be linearly represented by its corresponding near neighbor points, thereby realizing secondary dimension reduction of the vibration signal feature and obtaining an observation sequence. By performing secondary dimension reduction on the vibration signal feature based on the preset local linear embedding rule, the local features of the tool wear state can be well preserved. It should be noted that the tool wear state usually has obvious local features and manifold structure, and the above local linear embedding rule can preserve these local features and manifold structure while reducing dimensions, thereby better reflecting the changes of the tool wear state. Moreover, the vibration signal feature after secondary dimension reduction, that is, the observation sequence, is easier to classify and identify. By mapping high-dimensional data into a low-dimensional space while preserving the local features and manifold structure of the original data, the features after dimension reduction are easier to classify and identify. The recognition accuracy and reliability of the tool wear state are improved. The influence of noise and outliers on tool wear state recognition is reduced, and the robustness of tool wear state recognition is improved.

[0064] S104: inputting the observation sequence into a pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction, and obtaining a recognition result and a prediction result.

[0065] Specifically, by inputting the observation sequence into the pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction, the tool wear state can be accurately distinguished, and the tool wear state can be predicted.

[0066] S105: feeding back the recognition result and the prediction result to an associated terminal device. The associated terminal device can be a pre-associated terminal device, such as a computer, a mobile phone, etc. By feeding back the recognition result and the prediction result to the associated terminal device, relevant personnel can monitor the tool wear state in real time.

[0067] Please refer to Figure 2 In some embodiments, the step of inputting the tool wear vibration signal into a preset convolutional neural network to perform feature extraction and primary dimension reduction to obtain a vibration signal feature includes:

[0068] S201: inputting the tool wear vibration signal into a convolutional layer of the convolutional neural network to perform feature extraction and obtain a target feature. By inputting the tool wear vibration signal into the convolutional layer of the convolutional neural network to perform feature extraction, automatic extraction of the tool wear vibration signal feature can be realized, the accuracy is high, and problems such as poor recognition effect caused by manual feature extraction are avoided.

[0069] In some embodiments, the mathematical expression of the convolutional layer performing feature extraction is:

[0070]

[0071] wherein f represents an activation function, represents a convolution kernel, represents the mth feature map in the lth layer, represents a preset bias value, represents the extracted target feature.

[0072] S202: inputting the target feature into a pooling layer of the convolutional neural network to perform one-dimensional reduction to obtain the vibration signal feature. Specifically, the target feature is inputted into the pooling layer, the target feature outputted by the convolutional layer is processed by using a maximum pooling method, the maximum statistical value of each pooling unit is selected to represent the corresponding feature, and thus the vibration signal feature is obtained. In some embodiments, the mathematical expression of the pooling processing is as follows:

[0073]

[0074] wherein, represents the vibration signal feature after two-dimensional reduction, and max represents the maximum statistical value, represents the tth neuron in the hth channel in the lth layer, and T represents a pooling step.

[0075] Please refer to Figure 3 In some embodiments, based on a preset local linear embedding rule, the vibration signal feature is subjected to two-dimensional reduction, and the step of obtaining the observation sequence comprises:

[0076] S301: determining initial neighbor points of each sample point in the vibration signal feature based on a preset neighbor point discrimination sub-rule, each sample point corresponding to one or more initial neighbor points. The neighbor point discrimination sub-rule can be set according to actual conditions, for example, the number of neighbor points is set to 3, 5, etc.

[0077] S302: sorting and screening the initial neighbor points according to the Euclidean distance between the sample points and the corresponding initial neighbor points to obtain at least one target neighbor point. That is, the initial neighbor points of the sample points are sorted according to the Euclidean distance between the sample points and the corresponding initial neighbor points, and a preset number of neighbor points in the front of the sorting are selected as the target neighbor points.

[0078] S303: Obtain a local reconstruction weight matrix of the sample point corresponding to the target neighbor point by fitting a linear relationship to the target neighbor point. That is, by expressing the relationship between the sample point and its target neighbor point as a linear relationship, the local reconstruction weight matrix of the sample point is obtained. The local reconstruction weight matrix can be obtained by using the existing local linear embedding algorithm (LLE, Locally Linear Embedding), which will not be described here.

[0079] S304: Obtain the observed sequence after secondary dimension reduction based on the local reconstruction weight matrix of each sample point and its target neighbor point. That is, by using the local reconstruction weight matrix and the target neighbor point, the sample point is mapped to a preset embedding coordinate for secondary dimension reduction, and the finally obtained data is taken as the observed sequence.

[0080] For the implementation of local linear embedding, please refer to Figure 4 First, select the neighbor, that is, select the initial neighbor point of the sample point X i , and sort and filter the initial neighbor point according to the Euclidean distance between the sample point X i and the corresponding initial neighbor point, to obtain at least one target neighbor point. Second, linear reconstruction, that is, by fitting a linear relationship to the target neighbor point (X k , X j ), a local reconstruction weight matrix of the sample point corresponding to the target neighbor point is obtained, Figure 4 W ik and W ij in i respectively represent the weight between the sample point X k and the target neighbor point (X j , X i ). Then, map to the embedding coordinate, that is, based on the local reconstruction weight matrix of each sample point and its target neighbor point, the sample point X i is mapped to a preset embedding coordinate Y k , and the target neighbor point X j , X k is mapped to Y j , to obtain the observed sequence after secondary dimension reduction.

[0081] Please refer to Figure 5 In some embodiments, the step of obtaining the hidden Markov model comprises:

[0082] S501: Obtain an observed sequence sample set, which includes one or more observed sequence samples. Wherein, the observed sequence sample is like O = {O1, O2,... O T}, O TThe feature after the secondary dimension reduction is represented. By obtaining the observation sequence sample set, subsequent model training is facilitated.

[0083] S502: The parameters of the preset original model are initialized, and the number of hidden states is set, the hidden states including initial wear, intermediate wear and severe wear.

[0084] It should be noted that the original model is an original hidden Markov model λ, and the initial parameters thereof include an initial state probability matrix Л, a state transition probability matrix A and an emission matrix B. The hidden states can be set according to actual needs, such as normal and complete damage, and details are not described herein.

[0085] S503: The observation sequence sample is input into the initialized original model, a preset forward-backward algorithm is used to obtain a first intermediate quantity and a second intermediate quantity, the first intermediate quantity being obtained based on the forward probability and the backward probability of each hidden state of the observation sequence sample, and the second intermediate quantity being obtained based on the forward probability, the backward probability and the state transition probability of each hidden state of the observation sequence sample.

[0086] Specifically, the mathematical expression of the first intermediate quantity is:

[0087]

[0088] wherein, γ t (i) represents the first intermediate quantity, P(i t =q i , O; λ) represents the probability of the observation sequence sample of the hidden state q i , i t represents the hidden state, P(O; λ) represents the output probability of the observation sequence sample, a t (i) represents the recursively obtained forward probability, and β t (i) represents the recursively obtained backward probability.

[0089] The mathematical expression of the second intermediate quantity is:

[0090]

[0091] wherein, ξ t (i,j) represents the second intermediate quantity, P(i t =q i , i t+1 =q j , O; λ) represents the probability of the observation sequence sample of the hidden state q i at the time t and the hidden state q j at the time t+1, a ij represents the state transition probability from the hidden state q iConvert to hidden state q j Probability of occurrence, b j (O t+1 ) represents the feature O t+1 is the hidden state q j The emission matrix corresponding to the hidden state β t+1 (j) represents the backward probability obtained by recursion at time t+1.

[0092] S504: update and iterate the model parameters of the original model based on the first intermediate quantity and the second intermediate quantity, to obtain the trained hidden Markov model.

[0093] Please refer to Figure 6 In some embodiments, the observation sequence is input into the pre-trained hidden Markov model to perform tool wear state recognition, and the step of obtaining the recognition result includes:

[0094] S601: input the observation sequence into the hidden Markov model to perform initial forward probability calculation and / or initial backward probability calculation, and obtain the initial forward probability and / or initial backward probability of the observation sequence.

[0095] Specifically, given a hidden Markov model λ=(A, B, П), A is a state transition probability matrix, B is an emission matrix, and П is an initial state probability matrix. Given an observation sequence O={O1, O2,...O T}.

[0096] The step of performing initial forward probability calculation includes: inputting the observation sequence into the hidden Markov model, calculating the forward probability of each hidden state at the initial time, to obtain:

[0097] α1(i)=π i b i (O1),i=1,2,…N

[0098] Wherein, α1(i) represents the initial forward probability of the i-th hidden state at time 1, π i represents the initial state probability matrix of the i-th hidden state, b i (o1) represents the emission matrix of feature O1 as the i-th hidden state, and N represents the number of hidden states.

[0099] Further, the step of performing initial backward probability calculation includes: inputting the observation sequence into the hidden Markov model, calculating the backward probability of each hidden state at the initial time, to obtain:

[0100] β T (i)=1

[0101] Wherein, β T(i) represents the initial backward probability of the i-th hidden state at time T.

[0102] It can be understood that the initial time of the forward probability calculation is usually time 1, and the initial time of the backward probability calculation is usually time T, i.e., the last time.

[0103] S602: forwardly recursively calculating the initial forward probability according to time sequence to obtain a target forward probability, and / or backwardly recursively calculating the initial backward probability according to time sequence to obtain a target backward probability.

[0104] Specifically, the initial forward probability is forwardly recursively calculated according to time sequence, i.e., the forward probabilities at times 2, 3, … T are recursively calculated to obtain the target forward probability. The mathematical expression of the step of forwardly recursively calculating the initial forward probability according to time sequence is as follows:

[0105]

[0106] wherein, α t+1 (i) represents the probability obtained by forward recursion, α t (j) represents the probability that the hidden state is q j at time t, a ji represents the probability of transition from the hidden state q j to the hidden state q i , i.e., the probability of transition from the j-th hidden state to the i-th hidden state, b i (O t+1 ) represents the emission matrix of the feature O t+1 to the i-th hidden state.

[0107] Further, the initial backward probability is backwardly recursively calculated according to time sequence, i.e., the backward probabilities at times T-1, T-2, … 2 are recursively calculated to obtain the target backward probability. The mathematical expression of the target backward probability is as follows:

[0108]

[0109] wherein, β t (i) represents the target backward probability obtained by backward recursion, b j (O t+1 ) represents the emission matrix of the feature O t+1 to the j-th hidden state, β t+1 (j) represents the backward probability of the j-th hidden state at time t+1.

[0110] S603: obtaining the recognition result based on the target forward probability and / or the target backward probability.

[0111] Specifically, the mathematical expression of obtaining the recognition result based on the target forward probability is as follows:

[0112] N

[0113] P(O|λ)=Σα T (i)

[0114] i=1

[0115] wherein, P(O|λ) represents a recognition result, i.e. an observation sequence probability, α T (i) represents a target forward probability obtained by forward recursion, i.e. a probability that the i-th hidden state at time T.

[0116] Further, based on the target backward probability, a mathematical expression for obtaining the recognition result is:

[0117]

[0118] wherein, π i represents an initial state probability matrix corresponding to the i-th hidden state, b i (O1) represents an emission matrix of feature O1 being the i-th hidden state, and β1(i) represents a target backward probability obtained by backward recursion, i.e. a probability that the i-th hidden state at time 1.

[0119] In addition, the recognition result obtained based on the target forward probability and the target backward probability is usually the same, and therefore, when the forward probability calculation and the backward probability calculation are performed simultaneously, either the recognition result obtained by the forward probability calculation or the recognition result obtained by the backward probability calculation can be taken as the final recognition result.

[0120] Please refer to Figure 7 In some embodiments, the observation sequence is input into a pre-trained hidden Markov model to perform tool wear state prediction, and the step of obtaining a prediction result includes:

[0121] S701: outputting the observation sequence to the hidden Markov model to obtain initialized local state data.

[0122] Specifically, the observation sequence O={O1,O2,...O T} is input into a hidden Markov model λ=(A,B,Л) to perform local state initialization to obtain initialized local state data, and a mathematical expression thereof is:

[0123] δ1(i)=π i b i (O1),i=1,2…N

[0124] Ψ1(i)=0,i=1,2…N

[0125] Here, δ1(i) and Ψ1(i) represent two state parameters for initializing local state data. δ1(i) focuses on the initial probability, while Ψ1(i) focuses on the transition probability between hidden states.

[0126] S702: Based on the initialized local state data, perform local state dynamic planning to obtain the predicted local state at multiple time points.

[0127] Specifically, based on the initialized local state data, local state dynamic programming is performed to obtain the predicted local state at times t = 2, 3, ... T. The mathematical expression of the predicted local state is as follows:

[0128]

[0129]

[0130] Where, δ t (i), Ψ t (i) represents two state parameters for predicting the local state, δ t-1 (j) represents the value of δ corresponding to the j-th hidden state at time t-1, b i (O t ) represents feature O t Let be the emission matrix for the i-th hidden state.

[0131] S703: Based on the predicted local state, obtain the target hidden state sequence probability and the target hidden state sequence at the target time.

[0132] In some embodiments, the mathematical expression for the probability of the target hidden state sequence is:

[0133]

[0134] The mathematical expression for the target hidden state sequence is:

[0135]

[0136] Among them, P * Represents the probability of the target's hidden state sequence. Denotes the target hidden state sequence, δ T (i) represents the δ value corresponding to the i-th hidden state at time T.

[0137] S704: Based on the predicted local state and the target hidden state sequence, backtracking is performed to obtain the final hidden state sequence; the final hidden state sequence is used as the prediction result. That is, using Ψ t (i) Backtracking through time points t = T-1, T-2, ..., 1, the final hidden state sequence is obtained, and the mathematical expression of the final hidden state sequence is:

[0138]

[0139] wherein, represents the final hidden state sequence.

[0140] By using the trained hidden Markov model to predict the tool wear state in the above steps, the future wear state and wear degree of the tool can be better evaluated and controlled by relevant personnel, and the implementability is relatively strong. The accuracy of the final hidden state sequence predicted by the above method is relatively high.

[0141] In some embodiments, the step of feeding back the identification result and the prediction result to the associated terminal device comprises: issuing an alarm or generating warning information based on the identification result and the prediction result; and feeding back the warning information to the associated terminal device.

[0142] The tool wear state identification system provided by the present application is described below. The tool wear state identification system described below can be referred to in correspondence with the tool wear state identification method described above.

[0143] Please refer to Figure 8 The tool wear state identification system provided by the present embodiment comprises:

[0144] The acquisition module 801 is configured to acquire a tool wear vibration signal.

[0145] The convolution module 802 is configured to input the tool wear vibration signal into a preset convolutional neural network to perform feature extraction and one-dimensional dimension reduction, and obtain a vibration signal feature.

[0146] The dimension reduction module 803 is configured to perform two-dimensional dimension reduction on the vibration signal feature based on a preset local linear embedding rule, and obtain an observation sequence.

[0147] The identification and prediction module 804 is configured to input the observation sequence into a pre-trained hidden Markov model to perform tool wear state identification and tool wear state prediction, and obtain an identification result and a prediction result.

[0148] The communication module 805 is configured to feed back the identification result and the prediction result to an associated terminal device. The acquisition module 801, the convolution module 802, the dimension reduction module 803, the identification and prediction module 804, and the communication module 805 are connected. The tool wear state identification system provided by the present embodiment can automatically realize feature extraction of the tool wear vibration signal, has relatively high accuracy and relatively strong real-time performance. Moreover, the tool wear state identification accuracy of the system is relatively high, the tool wear state can be accurately predicted, the implementability is relatively strong, and the cost is relatively low.

[0149] In some embodiments, the convolution module 802 inputs the tool wear vibration signal into a preset convolutional neural network for feature extraction and first dimension reduction, and the step of obtaining a vibration signal feature includes:

[0150] The tool wear vibration signal is input into the convolutional layer of the convolutional neural network for feature extraction to obtain a target feature.

[0151] The target feature is input into the pooling layer of the convolutional neural network for first dimension reduction to obtain the vibration signal feature.

[0152] In some embodiments, the dimension reduction module 803 performs second dimension reduction on the vibration signal feature based on a preset local linear embedding rule, and the step of obtaining an observation sequence includes:

[0153] Based on a preset neighbor point discrimination sub-rule, the initial neighbor points of each sample point in the vibration signal feature are determined, and each sample point corresponds to one or more initial neighbor points.

[0154] According to the Euclidean distance between the sample point and the corresponding initial neighbor point, the initial neighbor points are sorted and filtered to obtain at least one target neighbor point.

[0155] By fitting the linear relationship of the target neighbor points, a local reconstruction weight matrix of the sample point corresponding to the target neighbor point is obtained.

[0156] Based on the local reconstruction weight matrix of each sample point and its target neighbor point, the observation sequence after second dimension reduction is obtained.

[0157] In some embodiments, the step of obtaining the hidden Markov model includes:

[0158] An observation sequence sample set is obtained, which includes one or more observation sequence samples.

[0159] The parameters of a preset original model are initialized, and the number of hidden states is set, including initial wear, intermediate wear, and severe wear.

[0160] The observation sequence sample is input into the initialized original model, and a preset forward-backward algorithm is used to obtain a first intermediate quantity and a second intermediate quantity, the first intermediate quantity is based on the forward probability and the backward probability of each hidden state of the observation sequence sample, and the second intermediate quantity is based on the forward probability, the backward probability and the state transition probability of each hidden state of the observation sequence sample.

[0161] Based on the first intermediate quantity and the second intermediate quantity, the model parameters of the original model are updated and iterated to obtain the trained hidden Markov model.

[0162] In some embodiments, the step of inputting the observation sequence into the pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction to obtain the recognition result and the prediction result by the recognition and prediction module 804 comprises:

[0163] inputting the observation sequence into the hidden Markov model to perform initial forward probability calculation and / or initial backward probability calculation to obtain the initial forward probability and / or the initial backward probability of the observation sequence.

[0164] performing forward recursion on the initial forward probability in time sequence to obtain a target forward probability, and / or performing backward recursion on the initial backward probability in time sequence to obtain a target backward probability.

[0165] obtaining the recognition result based on the target forward probability and / or the target backward probability.

[0166] In some embodiments, the step of inputting the observation sequence into the pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction to obtain the recognition result and the prediction result by the recognition and prediction module 804 comprises:

[0167] outputting the observation sequence from the hidden Markov model to obtain initialization local state data.

[0168] performing local state dynamic programming based on the initialization local state data to obtain a predicted local state at multiple time points.

[0169] obtaining a target hidden state sequence probability and a target hidden state sequence at a target time point based on the predicted local state.

[0170] performing backtracking based on the predicted local state and the target hidden state sequence to obtain a final hidden state sequence; taking the final hidden state sequence as the prediction result.

[0171] In some embodiments, the step of feeding back the recognition result and the prediction result to the associated terminal device by the communication module 805 comprises:

[0172] issuing an alarm or generating warning information based on the recognition result and the prediction result.

[0173] feeding back the warning information to the associated terminal device.

[0174] Figure 9 An example of an entity structure diagram of an electronic device is shown in FIG. 1. Figure 9As shown, the electronic device can include a processor 910, a communications interface 920, a memory 930, and a communications bus 940, wherein the processor 910, the communications interface 920, and the memory 930 complete mutual communication through the communications bus 940. The processor 910 can invoke a logical instruction in the memory 930 to execute a tool wear state identification method, which includes: collecting a tool wear vibration signal; inputting the tool wear vibration signal into a preset convolutional neural network to perform feature extraction and first dimension reduction, and obtaining a vibration signal feature; based on a preset local linear embedding rule, performing second dimension reduction on the vibration signal feature to obtain an observation sequence; inputting the observation sequence into a pre-trained hidden Markov model to perform tool wear state identification and tool wear state prediction, and obtaining an identification result and a prediction result; and feeding back the identification result and the prediction result to an associated terminal device.

[0175] In addition, the logical instruction in the memory 930 described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application or parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0176] On the other hand, the present application also provides a computer program product, which includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the tool wear state identification method provided by the above-mentioned methods, which includes: collecting a tool wear vibration signal; inputting the tool wear vibration signal into a preset convolutional neural network to perform feature extraction and first dimension reduction, and obtaining a vibration signal feature; based on a preset local linear embedding rule, performing second dimension reduction on the vibration signal feature to obtain an observation sequence; inputting the observation sequence into a pre-trained hidden Markov model to perform tool wear state identification and tool wear state prediction, and obtaining an identification result and a prediction result; and feeding back the identification result and the prediction result to an associated terminal device.

[0177] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a tool wear state recognition method provided by any of the above methods, the method comprising: collecting a tool wear vibration signal; inputting the tool wear vibration signal into a preset convolutional neural network to perform feature extraction and first dimension reduction, and obtaining a vibration signal feature; performing second dimension reduction on the vibration signal feature based on a preset local linear embedding rule, and obtaining an observation sequence; inputting the observation sequence into a pre-trained hidden Markov model to perform tool wear state recognition and tool wear state prediction, and obtaining a recognition result and a prediction result; and feeding back the recognition result and the prediction result to an associated terminal device.

[0178] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0179] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software plus a necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0180] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying the wear condition of a cutting tool, characterized in that, include: Collect vibration signals from tool wear; The tool wear vibration signal is input into a preset convolutional neural network for feature extraction and one dimensionality reduction to obtain vibration signal features; Based on a preset local linear embedding rule, the vibration signal features are subjected to secondary dimensionality reduction to obtain the observation sequence; The observation sequence is input into a pre-trained Hidden Markov Model to identify and predict tool wear status, and the identification and prediction results are obtained. The identification and prediction results are fed back to the associated terminal device; The step of performing a second-order dimensionality reduction on the vibration signal features based on a preset local linear embedding rule to obtain the observation sequence includes: Based on the preset nearest neighbor discrimination sub-rule, the initial nearest neighbor of each sample point in the vibration signal feature is determined, and each sample point corresponds to one or more initial nearest neighbor points; Based on the Euclidean distance between the sample point and its corresponding initial nearest neighbor, the initial nearest neighbor is sorted and filtered to obtain at least one target nearest neighbor. By fitting a linear relationship between the target nearest neighbors, the local reconstruction weight matrix of the sample points corresponding to the target nearest neighbors is obtained; Based on the local reconstruction weight matrix of each sample point and its target nearest neighbor, the observation sequence after secondary dimensionality reduction is obtained. The steps for obtaining the Hidden Markov Model include: Obtain an observation sequence sample set, which includes one or more observation sequence samples; The parameters of the preset original model are initialized, and the number of hidden states is set, including: initial wear, intermediate wear, and severe wear; The observed sequence samples are input into the initialized original model, and a first intermediate quantity and a second intermediate quantity are obtained using a preset forward-backward algorithm. The first intermediate quantity is obtained based on the forward probability and backward probability of each hidden state of the observed sequence samples, and the second intermediate quantity is obtained based on the forward probability, backward probability and state transition probability of each hidden state of the observed sequence samples. Based on the first intermediate value and the second intermediate value, the model parameters of the original model are updated and iterated to obtain the trained Hidden Markov Model.

2. The tool wear condition identification method according to claim 1, characterized in that, The steps of inputting the tool wear vibration signal into a preset convolutional neural network for feature extraction and dimensionality reduction to obtain vibration signal features include: The tool wear vibration signal is input into the convolutional layer of the convolutional neural network for feature extraction to obtain target features; The target features are input into the pooling layer of the convolutional neural network for dimensionality reduction to obtain the vibration signal features.

3. The tool wear condition identification method according to claim 1, characterized in that, The steps of inputting the observed sequence into a pre-trained Hidden Markov Model to identify and predict tool wear status, and obtaining the identification and prediction results, include: The observation sequence is input into the Hidden Markov Model to perform initial forward probability calculation and / or initial backward probability calculation, thereby obtaining the initial forward probability and / or initial backward probability of the observation sequence. The initial forward probability is recursively calculated in the time sequence to obtain the target forward probability, and / or the initial backward probability is recursively calculated in the time sequence to obtain the target backward probability. The recognition result is obtained based on the target forward probability and / or target backward probability.

4. The tool wear condition identification method according to claim 1, characterized in that, The steps of inputting the observed sequence into a pre-trained Hidden Markov Model to identify and predict tool wear status, and obtaining the identification and prediction results, include: The observation sequence is input into the hidden Markov model to obtain the initial local state data; Based on the initial local state data, perform local state dynamic programming to obtain the predicted local state at multiple time points. Based on the predicted local state, obtain the target hidden state sequence probability and the target hidden state sequence at the target time. Based on the predicted local state and the target hidden state sequence, backtracking is performed to obtain the final hidden state sequence; the final hidden state sequence is used as the prediction result.

5. The tool wear condition identification method according to claim 1, characterized in that, The step of feeding back the identification and prediction results to the associated terminal device includes: Based on the identification and prediction results, an alarm is issued or a warning message is generated. The warning information is then sent back to the associated terminal device.

6. A tool wear condition identification system, characterized in that, include: The acquisition module is used to acquire vibration signals from tool wear. The convolution module is used to input the tool wear vibration signal into a preset convolutional neural network, perform feature extraction and one dimensionality reduction, and obtain vibration signal features; The dimensionality reduction module is used to perform secondary dimensionality reduction on the vibration signal features based on a preset local linear embedding rule to obtain the observation sequence; The identification and prediction module is used to input the observation sequence into a pre-trained Hidden Markov Model to identify and predict the tool wear state, and obtain the identification and prediction results. The communication module is used to feed back the identification results and prediction results to the associated terminal device; The device is also used for: Based on the preset nearest neighbor discrimination sub-rule, the initial nearest neighbor of each sample point in the vibration signal feature is determined, and each sample point corresponds to one or more initial nearest neighbor points; Based on the Euclidean distance between the sample point and its corresponding initial nearest neighbor, the initial nearest neighbor is sorted and filtered to obtain at least one target nearest neighbor. By fitting a linear relationship between the target nearest neighbors, the local reconstruction weight matrix of the sample points corresponding to the target nearest neighbors is obtained; Based on the local reconstruction weight matrix of each sample point and its target nearest neighbor, the observation sequence after secondary dimensionality reduction is obtained. The steps for obtaining the Hidden Markov Model include: Obtain an observation sequence sample set, which includes one or more observation sequence samples; The parameters of the preset original model are initialized, and the number of hidden states is set, including: initial wear, intermediate wear, and severe wear; The observed sequence samples are input into the initialized original model, and a first intermediate quantity and a second intermediate quantity are obtained using a preset forward-backward algorithm. The first intermediate quantity is obtained based on the forward probability and backward probability of each hidden state of the observed sequence samples, and the second intermediate quantity is obtained based on the forward probability, backward probability and state transition probability of each hidden state of the observed sequence samples. Based on the first intermediate value and the second intermediate value, the model parameters of the original model are updated and iterated to obtain the trained Hidden Markov Model.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the tool wear state identification method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the tool wear condition identification method as described in any one of claims 1 to 5.

Images