Real-time crown strip tension detection method based on Gaussian mixture-hidden Markov model

Abstract

The invention discloses a crown strip real-time tension detection method based on a Gaussian mixture-hidden Markov model, and a crown strip real-time tension detection scheme based on the model is constructed by analyzing the rotation speed of a machine head winding drum and the translation speed of a winding head. According to the method, an observable machine head winding speed parameter is used for speculating the tension state of a crown strip, a machine head winding speed parameter sample is extracted on the basis of data collected in the forming stage in the tire production process, an expectation maximization algorithm is written to train and verify a Gaussian mixture-hidden Markov model, and the tension state of the crown strip is calculated. And finally, a relatively accurate crown strip tension prediction result is obtained.

Description

Technical Field

[0001] The invention relates to a real-time tension detection method of a cap band based on a mixed Gaussian-hidden Markov model, and belongs to the field of online tension detection. Background Art

[0002] As an important component in the automotive industry, the quality of tires directly affects the performance, safety and service life of the vehicle. The tire production process is complex and highly dependent on refined control, among which the molding process is one of the key links. The molding process involves the precise combination and superposition of multiple raw materials such as rubber, steel wire, and fiber through mechanical means to finally form a tire with a specific structure. The stability and precise control of this process are crucial to ensure the final quality of the tire. During the molding process, slight changes in process parameters may lead to significant fluctuations in material arrangement or bonding properties, which in turn affects the overall performance of the tire. Therefore, how to effectively predict quality during the molding process has become a core issue in improving tire manufacturing quality.

[0003] The main goal of the tire building process is to produce a semi-finished tire with specific physical and mechanical properties by accurately combining multiple components such as carcass cord, sidewall rubber, wire layer, etc. This process usually consists of multiple stages, including the laying of the carcass cord layer, the bonding of the tread and sidewall, and the forming of the wire ring. Each stage involves complex mechanical operations and physical and chemical reactions, such as bonding, stretching, shaping, etc., which are strongly affected by process conditions (such as temperature, pressure, time, tension, etc.). Due to the highly nonlinear relationship between the materials of each component and the process parameters, the results of the building process often show great uncertainty. Any slight deviation in any process parameter may lead to changes in the performance of the molded tire, thereby affecting its key quality indicators such as wear resistance, grip and service life.

[0004] In traditional production lines, the quality control of the tire molding process mainly relies on the operator's experience and post-testing. Although this method can ensure product quality to a certain extent, it also has obvious shortcomings. First, post-testing belongs to lagging control, and it is impossible to adjust the process parameters in real time during the production process, which easily leads to the production of batches of defective products. Secondly, although the operator's experience is valuable, due to the complexity and variability of the process, it is difficult to cope with all possible situations based on experience alone. Therefore, establishing a quality prediction and control system based on real-time data is the key to achieving refined management of the tire molding process.

[0005] In the commonly used real-time tension detection algorithm for the crown band, when training the HMM (Hidden Markov Model), the determination of the number of mixed Gaussian models and the number of states is highly subjective. The number of mixed Gaussian models and the number of states are usually set based on experience. At the same time, these two parameters have a great influence on the recognition accuracy of the model. Summary of the invention

[0006] The purpose of the present invention is to address the problem of structural parameter setting in the algorithm based on the hidden Markov model (HMM) in the real-time tension detection of the crown strap, and propose a real-time tension detection method for the crown strap based on a mixed Gaussian-hidden Markov model, so as to overcome the problem that in the traditional HMM algorithm, the structural parameters of the model are usually set based on experience, but the empirical setting is often difficult to accurately reflect the actual situation, resulting in the final structure of the model cannot be fully optimized, affecting the effect and accuracy of tension detection.

[0007] To achieve the above object, the technical solution adopted by the present invention is: a real-time tension detection method of the cap band based on a mixed Gaussian-hidden Markov model, which comprises the following steps:

[0008] Step 1: Obtain the crown strip tension data and analyze and process it: By analyzing the rotation speed V of the winding drum r and the translation speed V of the winding head t The characteristics and laws in the tire forming stage are used as the observed state parameters, and the cap strip tension state is selected as the implicit state parameter;

[0009] Step 2: Establish a mixed Gaussian-hidden Markov model: Based on the data provided by the production site, the mixed Gaussian-hidden Markov model is trained using the expectation maximization algorithm to learn the parameters of the crown band real-time tension detection model obtained by the training. in, is the Gaussian distribution mean vector, is the Gaussian distribution covariance matrix, M is the transition probability matrix of discrete latent variables at adjacent moments, satisfying k is the current hidden state, k′ is the non-current hidden state; π is the initialization vector of the hidden variable state, satisfying

[0010] Step 3: Use the crown band tension data to verify the mixed Gaussian-hidden Markov model.

[0011] Furthermore, the step 1 comprises:

[0012] 1.1) Set the observation state parameters: the rotation speed V of the winding drum r and the translation speed V of the winding head t A total of 2 quantities are used as observation state parameters;

[0013] 1.2) Extraction and processing of cap strip tension: Data is collected from the molding stage of the tire production process and divided into two groups, one for training and the other for verification.

[0014] Furthermore, the step 2 comprises:

[0015] 2.1) Mixed Gaussian-Hidden Markov Model and Parameters: Given input data Output Data Observation data is recorded as T is the number of samples, D x is the input data sample dimension, D y is the output data sample dimension; assuming there are K categories, the discrete latent variable is the indicator variable satisfy The mixed Gaussian-Hidden Markov model contains four model parameters, denoted as is the Gaussian distribution mean vector; is the Gaussian distribution covariance matrix; M is the transition probability matrix of discrete latent variables at adjacent moments, satisfying k is the current hidden variable state, k′ is the non-current hidden variable state; π is the initialization vector of the hidden variable state, satisfying

[0016] Parameter Description:

[0017] The real-time tension of the crown band has three types: high, low and qualified, K=3;

[0018] The real-time tension state of the cap band is predicted by observing the information, and the number of observable states is represented by the parameter, D x =2;

[0019] M represents the matrix composed of the transition probabilities between the three hidden states of high, low and qualified, that is, the state transition probability matrix;

[0020] The rotation speed V of the winding drum r and the translation speed V of the winding head t The probability distribution of the constituent observations is modeled by a Gaussian distribution, which is given by Mean vector and covariance matrix To represent, and finally form the output probability matrix of the observation value;

[0021] π is the probability matrix of the initial cap band tension, which is initialized by random or uniform values ​​under the condition of satisfying the Markov chain, and π=[1 / 2 1 / 2] is obtained by using the mean method;

[0022] 2.2) The mixed Gaussian-hidden Markov model uses the expected maximum algorithm iterative solution method to train the model: write the expected maximum algorithm, batch import learning training samples to train the mixed Gaussian-hidden Markov model, and then learn the parameters of the crown band real-time tension detection model obtained by training

[0023] Furthermore, in step 3, the Viterbi algorithm is used to decode and identify the hidden state of the trained mixed Gaussian-hidden Markov model, and the maximum value of the probability values ​​corresponding to the three crown strap tension states is decoded to represent the real-time tension predicted at this time, and the recognition accuracy of the crown strap real-time tension data of the trained mixed Gaussian-hidden Markov model is obtained.

[0024] The beneficial effect of the present invention is that the method of the present invention constructs a real-time tension detection model of the crown strip based on the in-depth analysis of the relationship between the rotation speed of the winding drum and the translation speed of the winding head. In this method, the observable rotation speed V of the winding drum is used to detect the tension of the crown strip. r and the translation speed V of the winding head t As a key parameter, the tension of the crown strip is inferred through a mixed Gaussian-hidden Markov model. The model can capture the dynamic changes of the winding speed to achieve more accurate tension prediction. This method effectively improves the accuracy of real-time tension detection and shows good applicability in actual production. It can adapt to different winding speed settings and provides important technical support for quality control of tire manufacturing. This not only solves the limitations of traditional methods that rely on experience to set model parameters, but also realizes a data-driven automated tension detection solution, greatly improving the real-time and accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a diagram of the winding process of the crown band of the present invention.

[0026] Figure 2 It is a data diagram of the winding tension of the cap band of the present invention.

[0027] Figure 3 It is a comparison chart between the model prediction results of the present invention and the true value.

[0028] Figure 4 It is a curve diagram of the model training loss and validation loss of the present invention changing with the number of training rounds.

[0029] Figure 5 It is a flow chart of the overall technology of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] A real-time tension detection method for the crown strip based on a mixed Gaussian-hidden Markov model is used to realize the real-time prediction of the crown strip tension. By analyzing the rotation speed V of the winding drum, r and the translation speed V of the winding head t The characteristics and rules of the tire forming stage are used as observation state parameters, and the cap strip tension state is selected as the implicit state parameter. Thus, a model for online prediction of the cap strip real-time tension using the observation state is established. Figure 1 Diagram of the crown band winding process. r is the rotation speed of the winding drum and V t is the translation speed of the winding head.

[0032] 1) Acquisition and analysis of crown band tension data

[0033] The crown band winding tension data diagram is drawn based on the data provided by the production site. Figure 2 As shown in the figure, the data of the cap layer winding tension shows that the tension value fluctuates within a certain range, with the lowest value reaching 2 and the highest value reaching 47, and the average value is approximately between 15 and 20. This may be due to quality problems or manufacturing defects in the cap layer itself, resulting in abnormal values ​​or outliers in the data. This range of tension fluctuations may be acceptable for some systems, but if our goal is to achieve lean control of tire quality, then the impact of abnormal values ​​or outliers on model establishment should be minimized as much as possible.

[0034] 2) Processing of crown band tension data

[0035] The acquired crown band tension data are divided into two parts, one part is used to train the mixed Gaussian-hidden Markov model, and the other part is used to verify the effectiveness of the model.

[0036] 3) Establishment of prediction algorithm based on mixed Gaussian-hidden Markov model

[0037] In order to obtain the mixed Gaussian-hidden Markov model for real-time tension prediction of the crown band, it is necessary to train the mixed Gaussian-hidden Markov model based on the data provided by the production site and use the expectation maximum algorithm to obtain the model parameters

[0038]

[0039] 3.1) Mixed Gaussian-Hidden Markov Model and Parameters

[0040] The mixed Gaussian-hidden Markov model is a statistical model used to describe the unknown parameters of hidden states. The model contains two random processes, one of which is a hidden state that cannot be directly observed and is described by the state transition probability matrix, while the other random process can output observable symbols defined by the output probability. The core of its application is to infer the internal invisible state from the observable external sequence, that is, to determine the hidden state through the observed parameters.

[0041] During the tire molding process, the real-time tension state of the cap strip cannot be directly observed, but parameters such as the head winding speed and raw material formula in the production process are observable. Therefore, the key to predicting the cap strip tension lies in using these observable process parameters to infer the real-time tension state that cannot be directly observed. The dual random process characteristics of the mixed Gaussian-hidden Markov model are very consistent with the law of cap strip tension changes. Therefore, it is a reasonable and effective choice to build a cap strip real-time tension prediction model based on this model.

[0042] The mixed Gaussian-Hidden Markov model contains four model parameters, denoted as is the Gaussian distribution mean vector; is the Gaussian distribution covariance matrix; M is the transition probability matrix of discrete latent variables at adjacent moments, satisfying k is the current hidden state, k′ is the non-current hidden state; π is the initialization vector of the hidden variable state, satisfying

[0043] 3.2) Training of parameters of mixed Gaussian-Hidden Markov model

[0044] For the training of the mixed Gaussian-Hidden Markov model, the unsupervised learning method in the parameter learning problem is needed. For the mixed Gaussian-Hidden Markov model, it is usually necessary to first give a set of inaccurate parameters, and then adjust the model parameters through repeated iterations and step-by-step refinement, so that the parameters are finally stabilized at an acceptable accuracy. That is, the mixed Gaussian-Hidden Markov model uses the expected maximum algorithm iterative solution method to train the model. The mixed Gaussian-Hidden Markov model is established, and 5000 groups of learning and training samples are imported in batches, of which 4000 groups are used for training and 1000 groups are used for verification. The parameters of the crown band real-time tension prediction model obtained by learning and training are:

[0045] Gaussian distribution mean vector

[0046]

[0047]

[0048]

[0049] Gaussian distribution covariance matrix

[0050]

[0051]

[0052]

[0053] State transfer matrix M:

[0054]

[0055] Initial probability distribution π:

[0056]

[0057] 3.3) Verification of the Mixed Gaussian-Hidden Markov Model

[0058] The prediction of the tension of the crown band by the mixed Gaussian-hidden Markov model belongs to the decoding application of the mixed Gaussian-hidden Markov model. In order to verify the effectiveness of the model training, the Viterbi algorithm is used to decode and identify the hidden state and predict the tension of the crown band. In order to verify the correctness of the algorithm proposed in this paper, 1000 sets of data are used for testing and verification. The test results and performance indicators are shown in Table 1:

[0059] Table 1 Model prediction performance indicators

[0060] Mean square error Root mean square error Mean absolute error <![CDATA[R 2 ]]> 9.04762 3.00793 1.97238 0.71865

[0061] Comparison of predicted results with true values ​​can be seen in Figure 3 .

[0062] The curves of model training loss and validation loss changing with training rounds are shown in Figure 4 .

[0063] The present invention predicts the real-time tension of the crown band strip by establishing a real-time tension prediction model of the crown band strip based on the theory of the mixed Gaussian-hidden Markov model. In view of the need to adopt an unsupervised learning method when training the mixed Gaussian-hidden Markov model, this paper adopts the expectation maximization algorithm to iteratively learn the model parameters, thereby further improving the accuracy of the real-time tension prediction model of the crown band strip.

[0064] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the protection scope of the present invention in any form, and all technical solutions obtained by equivalent replacement and the like fall within the protection scope of the present invention. The parts not involved in the present invention are the same as the prior art or can be implemented by the prior art.

Claims

1. A real-time tension detection method for crown band based on mixed Gaussian-hidden Markov model, characterized in that: The steps include: Step 1: Obtain the crown strip tension data and analyze and process it: By analyzing the rotation speed V of the winding drum r and the translation speed V of the winding head t The characteristics and laws in the tire forming stage are used as the observed state parameters, and the cap strip tension state is selected as the implicit state parameter; Step 2: Establish a mixed Gaussian-hidden Markov model: Based on the data provided by the production site, the mixed Gaussian-hidden Markov model is trained using the expectation maximization algorithm to learn the parameters of the crown band real-time tension detection model obtained by the training. in, is the Gaussian distribution mean vector, is the Gaussian distribution covariance matrix, M is the transition probability matrix of discrete latent variables at adjacent moments, satisfying k is the current hidden state, k′ is the non-current hidden state; π is the initialization vector of the hidden variable state, satisfying Step 3: Use the crown band tension data to verify the mixed Gaussian-hidden Markov model.

2. The method for real-time tension detection of the girdle strip based on a mixed Gaussian-hidden Markov model according to claim 1, characterized in that: The step 1 comprises: 1.1) Set the observation state parameters: the rotation speed V of the winding drum r and the translation speed V of the winding head t A total of 2 quantities are used as observation state parameters; 1.2) Extraction and processing of cap strip tension: Data is collected from the molding stage of the tire production process and divided into two groups, one for training and the other for verification.

3. The method for real-time tension detection of the girdle strip based on a mixed Gaussian-hidden Markov model according to claim 1, characterized in that: The step 2 comprises: 2.1) Mixed Gaussian-Hidden Markov Model and Parameters: Given input data Output Data Observation data is recorded as T is the number of samples, D x is the input data sample dimension, D y is the output data sample dimension; assuming there are K categories, the discrete latent variable is the indicator variable satisfy The mixed Gaussian-Hidden Markov model contains four model parameters, denoted as is the Gaussian distribution mean vector; is the Gaussian distribution covariance matrix; M is the transition probability matrix of discrete latent variables at adjacent moments, satisfying k is the current hidden variable state, k′ is the non-current hidden variable state; π is the initialization vector of the hidden variable state, satisfying Parameter Description: The real-time tension of the crown band has three types: high, low and qualified, K=3; The real-time tension state of the cap band is predicted by observing the information, and the number of observable states is represented by the parameter, D x =2; M represents the matrix composed of the transition probabilities between the three hidden states of high, low and qualified, that is, the state transition probability matrix; The rotation speed V of the winding drum r and the translation speed V of the winding head t The probability distribution of the constituent observations is modeled by a Gaussian distribution, which is given by Mean vector and covariance matrix To represent, and finally form the output probability matrix of the observation value; π is the probability matrix of the initial cap band tension, which is initialized by random or uniform values ​​under the condition of satisfying the Markov chain, and π=[1 / 2 1 / 2] is obtained by using the mean method; 2.2) The mixed Gaussian-hidden Markov model uses the expected maximum algorithm iterative solution method to train the model: write the expected maximum algorithm, batch import learning training samples to train the mixed Gaussian-hidden Markov model, and then learn the parameters of the crown band real-time tension detection model obtained by training 4. The method for real-time tension detection of the girdle strip based on a mixed Gaussian-hidden Markov model according to claim 1, characterized in that: In step 3, the trained mixed Gaussian-hidden Markov model is decoded and identified using the Viterbi algorithm for hidden states, and the maximum value of the probability values ​​corresponding to the three crown strap tension states decoded represents the real-time tension predicted at this time, and the recognition accuracy of the crown strap real-time tension data of the trained mixed Gaussian-hidden Markov model is obtained.

Images