A tool state monitoring method of a hidden Markov model considering a degradation factor

Abstract

This invention discloses a tool condition monitoring method based on a Hidden Markov Model (HMM) considering degradation factors. This method can accurately diagnose and predict the wear state of machining tools during CNC machine tool operation and promptly prompt the machining system to replace the tool. The invention selects different forms of degradation factors based on the wear rate at different tool wear stages and adds them to the state transition matrix, making the HMM state transition matrix time-varying. This enhances the physical interpretability and physical consistency of the model during tool condition monitoring, adheres to the degradation law of tool wear, and improves the effectiveness of the tool wear condition monitoring model. This invention solves the problems of low physical interpretability and meaningless wear stage division in traditional tool monitoring models, enabling rapid and accurate monitoring of tool wear state and reducing losses caused by premature tool discarding or using over-worn tools that result in workpieces failing to meet accuracy requirements.

Description

Technical Field

[0001] This invention belongs to the field of mechanical processing and manufacturing, specifically relating to a tool condition monitoring method based on a hidden Markov model that considers degradation factors. Background Technology

[0002] In machining processes, cutting tools, as key components in modern precision CNC machine tool processing, are crucial factors in ensuring machining quality, improving production efficiency, and reducing production costs; however, they are also the most easily damaged components. Tool wear is inevitable during actual cutting, affecting machining efficiency, accuracy, and workpiece surface quality to some extent, and in severe cases, even damaging the machine tool. Studies have shown that tool failure downtime accounts for 20% of total machine tool downtime in milling. Online monitoring of tool wear during machining can save approximately 40% of tool costs. Therefore, accurate monitoring of tool wear is of great significance for ensuring machining accuracy, improving machining efficiency, and reducing production costs.

[0003] Although there is a lot of research on tool wear condition monitoring methods based on Hidden Markov Models, traditional Hidden Markov Models often assume that the tool state transition probability does not change with machining. They ignore the physical law that the tool state deteriorates continuously during actual machining, causing the tool wear state self-transition probability to decrease continuously, while the state transition probability in the direction of increased wear increases continuously. This leads to a decrease in the performance of the tool condition monitoring model. Summary of the Invention

[0004] This invention aims to address the shortcomings of existing technologies by proposing a tool wear state monitoring method based on a hidden Markov model that considers degradation factors. The method adds degradation factors to the state transition matrix, enabling the model to follow the degradation law of tool wear, thereby improving the effectiveness of the tool wear state monitoring model and achieving rapid and accurate monitoring of tool wear state.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0006] The tool condition monitoring method based on a hidden Markov model considering degradation factors, as described in this invention, is characterized by the following steps:

[0007] Step 1: Collect machining information from the CNC machine tool and acquire the milling force signals {F} of the cutting tool in three directions online using a three-axis force gauge mounted on the worktable. d (t)|d=x,y,z}, where F d (t) represents the milling force signal of the tool in the d-th direction during time period t; then, the tool wear information ω corresponding to time period t is obtained offline using an electron microscope.true (t) and used as the label value;

[0008] Step 2: Process the milling force signals {F} in three directions. d The milling force signals in the three directions are obtained by standardizing the values ​​of (t)d=x,y,z, and denoted as the milling force signals in the t-th time period. Thus, the milling force signal X from the beginning of time period t to time period (t+N) is obtained. t ={S t+1 ,S t+2 ,…,S t+i ,…,S t+N}, and X t ∈D, where D represents the tool wear condition monitoring dataset. S represents the standardized milling force signal of the tool in the d-th direction during time period t; t+i This represents the milling force signal during the (t+i)th time interval;

[0009] The continuous milling force signal X is obtained through signal processing methods. t Features are extracted to obtain a feature set; and the feature set is normalized to obtain a normalized feature set.

[0010] Step 3: Based on the tool wear information ω true (t), using equation (1) to establish a physical formula for tool wear, and using the tool wear curve fitted by the physical formula to expand the wear label so that each feature in the feature set corresponds to a wear label;

[0011] w(t)=A ln(Bt+1)+Ct 3 (1)

[0012] In equation (1), A, B, and C are three fitting coefficients, and t is the processing time;

[0013] Step 4: Construct a tool transition wear function based on the physical formula of tool wear, and use the tool transition wear function to divide the tool wear stages, thereby selecting different forms of degradation factors based on the differences between the divided wear stages;

[0014] Step 4.1: Based on the physical formula for tool wear, use equations (2) and (3) to obtain the transition wear function w of the convex part in time period t. E (t) and the concave transition wear function w at time t. L (t), where t is the processing time period;

[0015] w E (t)=A ln(Bt+1) (2)

[0016] w L (t)=Ct 3 (3)

[0017] Step 4.2, according to w E (t) and w L (t) divides the tool wear stages into three phases: initial wear stage, normal wear stage, and accelerated wear stage, including:

[0018] When w″(t) is less than 0, the tool is in the initial wear stage;

[0019] When w E (t)>w L (t),w′ E (t)=w′ L When (t), the tool condition transitions from the initial wear stage to the normal wear stage;

[0020] When w E (t)=w L (t),w′ E (t)<w′ L (t) or w(t)>w * At this time, the tool condition transitions from the normal wear stage to the accelerated wear stage; where w * This represents the maximum wear value on the flank face.

[0021] Step 4.3: Select different types of degradation factors based on the wear rate at different tool wear stages;

[0022] Step 5: Construct a hidden Markov model that considers degradation factors and use it for tool condition monitoring, including: time-varying state transition matrix module, observation probability matrix module, and model training module;

[0023] Step 5.1: The time-varying state transition matrix module establishes time-varying state transition matrices for different wear stages based on the degradation factor;

[0024] Step 5.2: The observation probability matrix module establishes the observation probability matrix based on the feature set using a Gaussian mixture model;

[0025] Step 5.3: The model training module trains the Hidden Markov Model considering degradation factors using the maximum likelihood function and optimizes the parameters in the Hidden Markov Model to obtain a trained tool wear state monitoring model.

[0026] Step 6: Collect milling force signals during the machining process in real time, process them according to the process in Step 2, and input them into the trained tool wear condition monitoring model to output the current tool wear condition.

[0027] Step 7: If the current tool wear condition reaches the critical tool wear state, stop the machine and change the tool; otherwise, return to step 6 to achieve real-time monitoring of the tool status.

[0028] The tool condition monitoring method based on a hidden Markov model considering degradation factors, as described in this invention, is also characterized in that step 4.3 includes:

[0029] The exponential degradation factor in the initial stage is obtained using equation (4):

[0030]

[0031] The constant degradation factor for the normal wear stage is obtained using equation (5):

[0032]

[0033] The exponential degradation factor for the accelerated wear stage is obtained using equation (6):

[0034]

[0035] In equations (6) and (7), a ii The self-transfer probability representing the tool wear state, a ij δ represents the probability that the tool wear state will shift towards the direction of accelerated wear; δ1 and δ2 represent the exponential degradation factors of the initial wear stage and the accelerated wear stage, respectively; σ represents the constant degradation factor; k represents the sequence number of the time period; and Δt represents the fixed interval between two observation times, i.e., the duration of each time period.

[0036] Step 5.1 includes:

[0037] Step 5.1.1: Obtain all hidden states in the Hidden Markov Model based on the tool wear process. First, divide the tool wear stage into 3 major stages according to Step 4. Then, use the K-means algorithm to divide each major stage into n minor stages, thus forming the hidden state set S = {s1, s2, ..., sn}. 3n}, where s i Let i represent the i-th hidden state of tool wear, where i∈[1,3n];

[0038] Step 5.1.2: Based on the number of hidden states, establish the initial state transition matrix for the corresponding dimension. Let a in A ij =P(q) t+1 =s j |q t =s i ) represents the tool state q in time period t. t In the i-th state s i The tool state q in the (t+1)th time periodt+1 Transition to the j-th state s j The probability of; let a ii =P(q) t+1 =s i |q t =s i ) represents the tool state q in time period t. t In the i-th state s i The tool state q in the (t+1)th time period t+1 Still in the i-th state s i The probability of i = 1, 2, ..., N, j = 1, 2, ..., N, where N represents the number of hidden states, N = 3n;

[0039] Step 5.1.3: Calculate the tool time-varying state transition matrix A1 for the initial wear stage in time period t, the tool time-varying state transition matrix A2 for the normal wear stage in time period t, and the tool time-varying state transition matrix A3 for the accelerated wear stage in time period t using equations (8), (9), and (10):

[0040]

[0041]

[0042]

[0043] Step 5.1.4: Use equation (11) to obtain the time-varying state transition matrix A for time period t. t :

[0044]

[0045] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the tool wear condition monitoring method, and the processor is configured to execute the program stored in the memory.

[0046] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the tool wear condition monitoring method.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] 1. This invention divides tool wear stages using an established physical formula for tool wear. First, it fits the physical formula with a small number of measured wear label values ​​to solve for the formula's parameters and expand the wear labels. Then, it uses the physical formula to calculate the wear rate and its derivative to further divide the wear stages. This method not only standardizes the division of tool wear stages under different working conditions but also makes the stage division independent of signal feature clustering, giving the divided tool wear stages greater physical meaning.

[0049] 2. This invention takes into account the shortcomings of traditional Hidden Markov Models in tool wear condition monitoring applications. By combining the degradation process of tool wear during machining, different forms of degradation factors are added to the state transition matrix of the Hidden Markov Model. This makes the model of this invention physically interpretable and improves the physical consistency of the model during tool condition monitoring, while making tool condition identification more accurate.

[0050] 3. This invention takes into account the signal inconsistency and wear rate difference at different wear stages, and proposes different forms of degradation factors for different degradation stages of the tool. The degradation factors are dynamically estimated based on real-time monitoring data, the state transition matrix at different times is calculated, and the wear state of the tool at the current stage is updated in real time, thereby improving the accuracy of tool monitoring. Attached Figure Description

[0051] Figure 1 This is a flowchart of the tool condition monitoring method based on a hidden Markov model that considers degradation factors, provided by the present invention.

[0052] Figure 2 This is a schematic diagram of the experimental platform provided by the present invention;

[0053] Figure 3 This is a schematic diagram illustrating the stages of wear state based on the physical formula for tool wear provided by this invention;

[0054] Figure 4 This is an algorithm structure diagram of the Hidden Markov Model that considers degradation factors provided by the present invention. Detailed Implementation

[0055] In this embodiment, a tool condition monitoring method based on a hidden Markov model considering degradation factors is described, such as... Figure 1 The following steps are shown:

[0056] Step 1: In this embodiment, the experimental platform is as follows: Figure 2 As shown, the machining data and tool wear information of the CNC machine tool are collected.

[0057] Step 1.1: Collect milling force signals from the cutting tool in three directions online using a force sensor mounted on the worktable.

[0058] {F d (t)|d=x,y,z}, where F d (t) represents the milling force signal of the tool in the d-th direction during time period t;

[0059] Step 1.2: Obtain tool wear information ω corresponding to time period t offline using an electron microscope. true (t) and used as the label value;

[0060] Step 2: Preprocess and extract features from the raw data obtained in Step 1;

[0061] Step 2.1: Milling force signals {F} in three directions d The milling force signals in the three directions are obtained by standardizing the data (t)|d=x,y,z}, denoted as the milling force signals in the time interval t. Thus, the milling force signal X from the start of time period t to time period t+N is obtained. t ={S t+1 ,S t+2 ,...,S t+i ,…,S t+N}, and X t ∈D, where D represents the tool wear condition monitoring dataset. S represents the standardized milling force signal of the tool in the d-th direction during time period t; t+i This represents the milling force signal during the (t+i)th time interval;

[0062] Step 2.2: Because the signals collected by the sensors cannot be directly input into the monitoring model, they need to be transformed from the continuous milling force signal X using short-time Fourier transform and wavelet transform. t Features are extracted from the sensor to obtain a feature set. In addition, since the dimensions or numerical scales of the multi-dimensional features extracted from the sensor are different, in order to avoid numerical overflow, improve the convergence speed of the algorithm, and reduce noise and interference, the extracted features are normalized to obtain a normalized feature set.

[0063] Step 3: Establish a physical formula based on tool wear according to the measured wear sample labels;

[0064] Step 3.1: Use the tool wear label ω measured with a LEICAMZ 12 microscope in Step 1. true (t), the fitting coefficients are obtained by fitting the physical formula of tool wear using MATLAB, as shown in equation (1):

[0065] w(t)=A ln(Bt+1)+Ct3 (1)

[0066] In equation (1), A, B, and C are three fitting coefficients, and t is the processing time;

[0067] Step 3.2: Expand the wear labels using the fitted physical formula for tool wear through interpolation, so that each set of features in the feature set corresponds to a wear label, and lay the theoretical foundation for the division of tool wear stages;

[0068] Step 4: In this embodiment, the wear stages are divided according to the physical formula of tool wear as shown in the diagram below. Figure 3 As shown, a tool transition wear function is constructed based on the physical formula of tool wear in step 3. The transition wear function is used to divide the tool wear stages, and different forms of degradation factors are selected based on the differences between the divided wear stages.

[0069] Step 4.1: Based on the physical formula for tool wear, use equations (2) and (3) to obtain the transition wear function w of the convex part in time period t. E (t) and concave transition wear function w L (t);

[0070] w E (t)=Aln(Bt+1) (2)

[0071] w L (t)=Ct 3 (3)

[0072] Step 4.2: Based on w E (t) and w L (t) divides the tool wear stages into three phases: initial wear stage, normal wear stage, and accelerated wear stage, including:

[0073] When w″(t) is less than 0, the tool is in the initial wear stage;

[0074] When w E (t)>w L (t),w′ E (t)=w′ L When (t), the tool condition transitions from the initial wear stage to the normal wear stage;

[0075] When w E (t)=w L (t),w′ E (t)<w′ L (t) or w(t)>w * At this time, the tool condition transitions from the normal wear stage to the accelerated wear stage; where w * This represents the maximum wear value on the flank face.

[0076] Previous studies lacked strict standards for classifying different tool wear stages, often relying on factors such as tool material, machining parameters, and the presence of coatings. However, most studies employed signal feature clustering or manual experience to classify tool states, lacking rigorous physical meaning. This invention, using a physical formula for tool wear, not only standardizes the classification of tool wear stages under different working conditions but also decouples the classification from signal feature clustering, making the resulting tool wear stages more physically meaningful.

[0077] Step 4.3: Select different types of degradation factors based on the wear rate at different tool wear stages:

[0078] The exponential degradation factor in the initial stage is obtained using equation (4):

[0079]

[0080] The constant degradation factor for the normal wear stage is obtained using equation (5):

[0081]

[0082] The exponential degradation factor for the accelerated wear stage is obtained using equation (6):

[0083]

[0084] Step 5: As Figure 4 As shown, a hidden Markov model considering degradation factors is constructed and trained for tool condition monitoring, including: a time-varying state transition matrix module, an observation probability matrix module, and a model training module;

[0085] Step 5.1.1: When constructing a Hidden Markov Model, the following are required: a set of hidden states, a set of observations, an initial state probability matrix, a time-varying state transition matrix, and an observation probability matrix.

[0086] Step 5.1.2: The set of hidden states S = {s1, s2, ..., s} 3n First, based on the physical formula of tool wear, the tool wear state is divided into 3 wear stages. Then, the K-means algorithm is used to further subdivide each major stage into n minor states. Therefore, the number of hidden states is 3n.

[0087] Step 5.1.3: Set of observations V = {v1, v2, ..., v} M}, where M represents the dimension of the observation, v iLet represent the observed value of the i-th feature. In this implementation case, the Pearson correlation coefficient is used to calculate the correlation between each feature in the feature set and the tool wear state. Features with strong correlation are selected as observed values ​​to construct the observation set.

[0088] Step 5.1.4: Initial state probability matrix π = {π1, π2, π3, ..., π} 3n}, where 3n represents the number of hidden states, i.e., the total number of divisions in the tool wear stage, and π i This represents the probability that the tool is initially in the i-th hidden state.

[0089] Step 5.1.5: Time-varying state transition matrix A t ={a ij (t)} N×N It changes as the processing progresses based on the initial state transition matrix, as shown in equation (10):

[0090]

[0091]

[0092]

[0093]

[0094] In the formula, A1, A2, and A3 are the time-varying state transition matrices of the tool during the initial wear stage, normal wear stage, and accelerated wear stage at time t, respectively, where a nn+1 (t) represents the tool state q in time period t. t In the nth state s n The tool state q in the (t+1)th time period t+1 Transition to the (n+1)th state s n+1 The probability of; and a 2n2n+1 (t) represents the tool state q in time period t. t In the 2nth state s 2n The tool state q in the (t+1)th time period t+1 Transition to the (2n+1)th state s 2n+1 The probability, a 2n2n+1 (t)=1-a 2n2n +(t / Δt)σ.

[0095] Step 5.2: Observation probability matrix B = {b} jk} N×M , where b jk (t)=P(v k |q t =sj ) indicates that at time t, the hidden state is in the j-th state, and the observed value is v. k The probability, where N represents the number of hidden states and M represents the dimension of the observations;

[0096] Step 5.3: After initializing the model parameters, train the model using the maximum likelihood function. Based on the observed values, estimate and optimize the parameters in the Hidden Markov Model considering the degradation factor. In this implementation case, first, the model parameters λ={π,A t The model is initialized with parameters B, δ1, δ2, σ. Then, based on the initial model parameters, the model parameters for each state are iteratively solved using the maximum likelihood algorithm, combined with the training data, until the model reaches its optimum, i.e., P(O|λ). n+1 )-P(O|λ n )≤ξ, where λ n+1 This is the result of the model solution, where ξ is usually taken as 10. -6 .

[0097] Step 6: Use the trained model library to monitor the tool wear condition and decide whether to stop the machine and replace the tool;

[0098] In this embodiment, the Viterbi algorithm is used to find the optimal hidden state i corresponding to each test sample observation sequence, i.e., i = arg max log(P(O|λ)). i This allows for the monitoring of tool wear conditions.

[0099] Step 7: If the wear threshold is reached, stop the machine and change the tool; otherwise, continue processing and return to step 6.

[0100] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the tool wear condition monitoring method, and the processor is configured to execute the program stored in the memory.

[0101] In this embodiment, a computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of a tool wear condition monitoring method.

[0102] In summary, this invention solves the problems of traditional tool wear monitoring methods that ignore the physical laws of tool wear degradation; it addresses the randomness and physical inconsistencies in the division of tool wear stages; it resolves the issue of decreased workpiece accuracy caused by failure to replace over-worn tools in a timely manner; and it eliminates the tool waste caused by premature tool replacement. This monitoring method offers advantages such as high diagnostic and monitoring accuracy, high physical consistency, convenient deployment, and strong adaptability.

Claims

1. A tool condition monitoring method using a hidden Markov model considering degradation factors, characterized in that, It is done according to the following steps: Step 1: Collect machining information from the CNC machine tool and collect milling force signals of the tool in three directions online using a three-axis force gauge installed on the worktable; Step 2: Standardize the milling force signals in the three directions; extract features from the continuous milling force signals using signal processing methods to obtain a feature set; and normalize the feature set to obtain a normalized feature set. Step 3: Based on the tool wear information, establish the physical formula for tool wear using equation (1); (1) In equation (1), , , There are three fitting coefficients. Processing time; Step 4: Construct a tool transition wear function based on the physical formula of tool wear, and use the tool transition wear function to divide the tool wear stages, thereby selecting different forms of degradation factors based on the differences between the divided wear stages; Step 4.1: Based on the physical formula for tool wear, obtain the transition wear function of the convex part in time period t using equations (2) and (3). and the concave transition wear function at time t ; (2) (3) Step 4.2, according to and The tool wear stages are divided into three phases: initial wear, normal wear, and accelerated wear. when When the value is less than 0, the tool is in the initial wear stage; when At that time, the tool condition transitions from the initial wear stage to the normal wear stage; when or At this time, the tool condition transitions from the normal wear stage to the accelerated wear stage; among which, This represents the maximum wear value on the flank face. Step 4.3: Select different types of degradation factors based on the wear rate at different tool wear stages; Step 5: Construct a hidden Markov model that considers degradation factors and use it for tool condition monitoring, including: time-varying state transition matrix module, observation probability matrix module, and model training module; Step 5.1: The time-varying state transition matrix module establishes time-varying state transition matrices for different wear stages based on the degradation factor; Step 5.2: The observation probability matrix module establishes the observation probability matrix using a Gaussian mixture model based on the normalized feature set. Step 5.3: The model training module trains the Hidden Markov Model considering degradation factors using the maximum likelihood function and optimizes the parameters in the Hidden Markov Model to obtain a trained tool wear state monitoring model. Step 6: Collect milling force signals during the machining process in real time, process them according to the process in Step 2, and input them into the trained tool wear condition monitoring model to output the current tool wear condition. Step 7: If the current tool wear condition reaches the critical tool wear state, stop the machine and change the tool; otherwise, return to step 6 to achieve real-time monitoring of the tool status.

2. The tool condition monitoring method using a hidden Markov model considering degradation factors according to claim 1, characterized in that, Step 4.3 includes: The exponential degradation factor in the initial stage is obtained using equation (4): (4) The constant degradation factor for the normal wear stage is obtained using equation (5): (5) The exponential degradation factor for the accelerated wear stage is obtained using equation (6): (6) In equations (4) to (6), The self-transfer probability representing the tool wear state. This represents the probability that the tool wear condition will shift towards a direction of increased wear. , These represent the exponential degradation factors for the initial wear stage and the accelerated wear stage, respectively. Represents the constant degradation factor; The sequence number representing the time period. This represents the fixed interval between two observation times, i.e., the duration of each time period.

3. The tool condition monitoring method using a hidden Markov model considering degradation factors according to claim 2, characterized in that, Step 5.1 includes: Step 5.1.1: Obtain all hidden states in the Hidden Markov Model based on the tool wear process. First, divide the tool wear stage into three major stages according to Step 4. Then, use the K-means algorithm to further divide each major stage into... These are small stages, thus forming a set of hidden states. ,in, The first indicator of tool wear A hidden state. ; Step 5.1.2: Based on the number of hidden states, establish the initial state transition matrix for the corresponding dimension. ,make middle Indicates the first Tool status during time period In the first each state And the first Tool status during time period Transformed into the first each state The probability of; let Indicates the first Tool status during time period In the first each state And the first Tool status during time period Still the first each state The probability, , , This indicates the number of hidden states. ; Step 5.1.3: Calculate the first step using equations (8), (9), and (10). Tool time-varying state transition matrix during the initial wear stage of the time period , No. Time-varying state transition matrix of the tool during the normal wear phase of a time period , No. Tool time-varying state transition matrix during the accelerated wear phase of the time period : (8) (9) (10) Step 5.1.4: Use equation (11) to obtain the first... Time-varying state transition matrix for a given period : (11)。 4. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the tool condition monitoring method of claim 1, 2, or 3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by the processor to perform the steps of the tool condition monitoring method according to claim 1, 2 or 3.

Images