Refractory material production line control method based on artificial intelligence

Abstract

The invention discloses a refractory material production line control method based on artificial intelligence, and relates to the technical field of automatic control, and the method comprises the steps: collecting the multi-dimensional production state and environment information of forming pressure, temperature, operation speed and humidity in real time, forming a data set PRM, and generating a high-quality feature data set FTS. The deep convolutional neural network technology is utilized to train and analyze the FTS, a predicted quality index QTY (t) is generated, the material quality state is accurately judged through matching with a material production quality state threshold CZ, and then a corresponding production control decision CTL (t) is triggered. And obtaining a production line regulation and control real-time evaluation coefficient EXE (t) at the time t through evaluation of a fixed period and a real-time feedback mechanism, and dynamically adjusting production parameters according to an evaluation result to form a closed-loop control system. The limitation of single parameter detection and simple comparison of a traditional method is overcome, and continuous optimization of production quality is achieved through loop iteration logic.

Description

Technical Field

[0001] The present invention relates to the field of automation control technology, and in particular to a refractory production line control method based on artificial intelligence. Background Art

[0002] In the invention of Chinese patent publication number CN118466420A, an intelligent control system and method for a refractory production line is disclosed, which relates to the field of intelligent control. It extracts features of the molding pressure values ​​at multiple predetermined time points within a predetermined time period to obtain molding pressure time series features, and extracts features of ultrasonic signals used to detect ceramic fiber blocks to obtain ultrasonic waveform features containing ceramic fiber block quality features, and then associates the above features to obtain the relationship between the quality of the ceramic fiber block and the molding pressure, and optimizes the associated features, and finally passes the optimized features through a classifier to obtain the control direction of the molding pressure. In this way, real-time control of the molding pressure of the ceramic fiber block is achieved, which helps to improve the automation level and product quality of the production line.

[0003] It can be seen that the testing and control is mainly focused on the single parameter of molding pressure, lacking comprehensive consideration of multiple dimensional parameters in the production line. Although this single parameter detection method can reflect some problems in production to a certain extent, it cannot fully capture the complex changes in each link of the production line, resulting in obvious limitations in dealing with multi-node problems in the production line. This solution only uses a simple pressure comparison method, which cannot effectively deal with other parameter changes that may affect material quality, resulting in blind spots in material production and quality control, which can easily lead to potential problems in the production process not being discovered and resolved in a timely manner. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a refractory production line control method based on artificial intelligence, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A refractory production line control method based on artificial intelligence, comprising the following steps:

[0006] S1. Several sensors are set up in the production line operation area to collect production line operation status information and environmental status information to form a collection data group PRM, and the collection data group PRM is transmitted in real time through a wireless network and Ethernet;

[0007] S2, performing data preprocessing and data feature extraction on the received collected data group PRM, including denoising preprocessing, filtering preprocessing and principal component analysis, to form a feature data set FTS;

[0008] S3. Establish a quality prediction learning model by using deep convolutional neural network technology, and input the acquired feature data set FTS as the input data set into the quality prediction learning model for training and analysis, obtain the predicted quality index QTY(t), and match it with the preset material production quality status threshold CZ to obtain the material quality status result;

[0009] S4. Triggering production control decision generation according to the material quality status result, obtaining the production control data group CTL(t) at time t, and regulating the production line generation control parameters according to the production control data group CTL(t);

[0010] S5. Evaluate the status of the control result of the production line output materials after regulation at a fixed period, obtain the real-time evaluation coefficient EXE(t) of the production line control at time t, match it with the preset post-regulation evaluation threshold TZ, and obtain the quality status of the material output after the production line regulation for iterative feedback.

[0011] Preferably, S1 further includes S11 and S12;

[0012] S11. Several sensors are set in the production line operation area to collect production line operation status information and environmental status information, including molding pressure information, equipment temperature information, operation speed information, environmental humidity information and equipment vibration frequency information, to form a time series data group, including molding pressure Pf(t), equipment temperature Te(t), equipment operation speed Vr(t), environmental humidity He(t) and equipment vibration Vu(t) at time t, which are integrated to form a collection data group PRM at time t, specifically PRM={Pf(t), Te(t), Vr(t), He(t), Vu(t)};

[0013] Among them, the production line operation area includes the raw material input area, molding area, heating area and cooling area; several sensors include temperature sensors, pressure sensors, humidity sensors and vibration sensors;

[0014] S12, the collected data group PRM at time t is transmitted to the data processing in real time by using the wireless network and Ethernet, and is classified and stored according to the data types included in the collected data group PRM.

[0015] Preferably, step S2 also includes S21 and S22;

[0016] S21, performing artificial intelligence data preprocessing on the received collected data group PRM, including denoising preprocessing and filtering preprocessing, wherein the denoising preprocessing includes using wavelet transform and mean filtering to perform denoising preprocessing, thereby eliminating high-frequency noise and random fluctuations in the collected data group PRM; the filtering preprocessing includes using a low-pass filter and a Kalman filter to perform filtering preprocessing, thereby eliminating related high-frequency noise in the collected data group PRM after the denoising preprocessing;

[0017] S22. Use principal component analysis to extract artificial intelligence information features from the preprocessed collected data group PRM to obtain the feature PCA1 of molding pressure, the feature PCA2 of equipment temperature, the feature PCA3 of equipment running speed, the feature PCA4 of ambient humidity and the feature PCA5 of equipment vibration to form a feature data set FTS, specifically FTS = {PCA1, PCA2, PCA3, PCA4, PCA5}.

[0018] Preferably, the feature PCA1 of the molding pressure is obtained by the following feature extraction method:

[0019] PCA1(t)=max(Pf(t))-min(Pf(t));

[0020] Wherein, PCA1(t) represents the characteristic value of the molding pressure at time t, specifically represents the fluctuation amplitude of the molding pressure, max(Pf(t)) represents the peak molding pressure at time t, and min(Pf(t)) represents the valley molding pressure at time t;

[0021] The PCA2 feature of the device temperature is obtained by the following feature extraction method:

[0022]

[0023] Where PCA2(t) represents the device temperature characteristic value at time t, n represents the total number of samples in a fixed period, and Te(t, i) represents the device temperature Te(t) at time t of sample i;

[0024] The feature PCA3 of the device running speed is obtained by the following feature extraction method:

[0025]

[0026] Where PCA3(t) represents the characteristic value of the equipment running speed at time t, Vr(t, i) represents the equipment speed at time t of sample i, and Indicates the average value of the device's operating speed;

[0027] The feature PCA4 of the ambient humidity is obtained by the following feature extraction method:

[0028] PCA4(t)=min(He(t));

[0029] Wherein, PCA4(t) represents the ambient humidity He(t) at time t, and min(He(t)) represents the valley value of the ambient humidity He(t) at time t;

[0030] The PCA5 feature of the device vibration is obtained by the following feature extraction method:

[0031] PCA5(t)=max(Vu(t));

[0032] Where PCA5(t) represents the equipment vibration characteristic value at time t, and max(Vu(t)) represents the peak value of the equipment vibration Vu(t) at time t.

[0033] Preferably, step S3 also includes S31 and S32;

[0034] S31. Establish a quality prediction learning model by using deep convolutional neural network technology, including multiple convolutional layers, pooling layers and fully connected layers, and input the acquired feature data set FTS as an input data set into the quality prediction learning model to perform artificial intelligence regulation of the quality prediction learning model training process, including using back propagation algorithm and optimization algorithm for regulation, and obtaining the predicted quality index QTY(t) through the analysis output of the output layer of the quality prediction learning model;

[0035] S32, matching the obtained predicted quality index QTY(t) with the preset material production quality status threshold CZ to obtain the material quality status result;

[0036] The material quality status results are obtained by the following matching method:

[0037]

[0038] When the return signal is 1, the material quality status result is qualified;

[0039] When the return signal is 0, the material quality status result is unqualified.

[0040] Preferably, the predicted quality index QTY(t) is obtained by the following calculation formula:

[0041] QTY(t)=fout(Wfc·fpool(Wconu·FTS(t)+bconu)+bfc);

[0042] Wherein, fout represents the output function, specifically represents the conversion of the output of the fully connected layer into the prediction quality index QTY(t), Wfc represents the weight matrix of the fully connected layer, fpool represents the pooling function, including the use of maximum pooling and average pooling to adjust the calculation amount of the model, Wconu represents the weight matrix of the convolutional layer, bconu represents the bias term of the convolutional layer, bfc represents the bias term of the fully connected layer, Wconu FTS(t)+bconu represents the calculation formula of the convolutional layer, specifically represents the extraction of local features from the input feature data FTS(t), and the adjustment of the expression of local features by adding the bias term bconu of the convolutional layer.

[0043] Preferably, step S4 also includes S41 and S42;

[0044] S41, triggering production control decision generation according to the material quality status result, and obtaining the production control data group CTL(t) at time t;

[0045] Among them, the generation of production control decisions is triggered specifically through the following mechanisms;

[0046] When the return signal of the material quality status result is 1, the material quality status result is qualified and does not trigger the generation of production control decision;

[0047] When the return signal of the material quality status result is 0, the material quality status result is unqualified, and the production control decision generation is triggered. By processing the molding pressure Pf(t), the equipment temperature Te(t) and the equipment running speed Vr(t), including proportionally adjusting the molding pressure Pf(t), the equipment temperature Te(t) and the equipment running speed Vr(t) in the form of percentage, the production control data group CTL(t) at time t is obtained;

[0048] S42. Generate control parameters for the production line according to the production control data group CTL(t). The control content includes the adjustment of molding pressure, the control of equipment temperature and the adjustment of production line running speed. The control content is specifically executed through PLC and DCS industrial control systems.

[0049] Preferably, the production control data set CTL(t) is generated by the following calculation formula:

[0050] CTL(t)={Pexec(t), Texec(t), Vexec(t)};

[0051] Wherein, Pexec(t) represents the molding pressure value after the production control is adjusted at time t, Texec(t) represents the equipment temperature value after the production control is adjusted at time t, and Vexec(t) represents the equipment operating speed value after the production control is adjusted at time t.

[0052] Preferably, step S5 also includes S51 and S52;

[0053] S51, the regulated production line output material is evaluated at a fixed period and the regulation result status is obtained to obtain the production line regulation real-time evaluation coefficient EXE(t) at time t;

[0054] The production line control real-time evaluation coefficient EXE(t) is obtained by the following calculation formula:

[0055] EXE(t)=δ·CTL(t)+ε·▽QTY(t);

[0056] In the formula, δ represents the control gain coefficient, ε represents the adjustment coefficient, and ▽QTY(t) represents the gradient matrix information of the predicted quality index QTY(t), which is specifically used to represent the trend of the predicted quality index QTY(t) within a fixed period;

[0057] S52. Match the preset post-control evaluation threshold TZ with the production line control real-time evaluation coefficient EXE(t) to obtain the material output quality status after the production line control for iterative feedback.

[0058] Preferably, the material output quality status after the production line is regulated is obtained by the following matching method:

[0059] When the real-time evaluation coefficient EXE(t) ≥ the post-regulation evaluation threshold TZ, the qualified evaluation result of the material output status after the production line regulation is obtained;

[0060] When the real-time evaluation coefficient EXE(t) is less than the post-regulation evaluation threshold TZ, the unqualified evaluation result of the material output status after the production line regulation is obtained, and steps S3, S4 and S5 are iteratively executed.

[0061] The present invention provides a refractory production line control method based on artificial intelligence, which has the following beneficial effects:

[0062] (1) Through steps S1 to S5, multiple sensors are deployed in the production line operation area to collect multi-dimensional production status and environmental information including molding pressure, temperature, running speed and humidity in real time to form a data group PRM, and after data preprocessing steps such as denoising, filtering and principal component analysis, a high-quality feature data set FTS is generated. The deep convolutional neural network technology is used to train and analyze FTS to generate a predicted quality index QTY(t), and the material quality status is accurately judged by matching it with the material production quality status threshold CZ, thereby triggering the corresponding production control decision CTL(t). Through a fixed-cycle evaluation and real-time feedback mechanism, the real-time evaluation coefficient EXE(t) of the production line control at time t is obtained, and the production parameters are dynamically adjusted according to the evaluation results to form a closed-loop control system. The integrated solution of multi-dimensional data acquisition and processing, multi-level intelligent analysis and dynamic feedback not only overcomes the limitations of single parameter detection and simple comparison of traditional methods, but also can more comprehensively and accurately control each link of the production process. It also realizes continuous optimization of production quality through cyclic iterative logic, significantly improving the stability and product quality of the refractory production line.

[0063] (2) By introducing multiple convolutional layers, pooling layers, and fully connected layers into the deep convolutional neural network model, the feature data set FTS is efficiently processed, and the model is continuously adjusted using the back-propagation algorithm and optimization algorithm to accurately generate the predicted quality index QTY(t). In addition, the scheme designs a matching mechanism based on QTY(t) and the preset material production quality status threshold CZ to ensure real-time monitoring and judgment of material quality. When the quality status is unqualified, the system automatically triggers the production control decision and makes refined adjustments to the molding pressure, equipment temperature, and operating speed. Through this mechanism of dynamically adjusting production parameters in percentage form, the scheme can flexibly respond to fluctuations in the production process, ensuring the stability of the production process and the continuous optimization of material quality. At the same time, through the PLC and DCS industrial control systems to execute the control content, the scheme realizes the intelligent management and automatic control of the production line, greatly improving production efficiency and quality consistency.

[0064] (3) By evaluating the production line after regulation at fixed periods, the real-time evaluation coefficient EXE(t) of production line regulation is generated, and the coefficient is matched with the preset post-regulation evaluation threshold TZ to obtain the quality status of the material after production line regulation in real time. In particular, the real-time evaluation coefficient EXE(t)(t) of production line regulation reflects the trend and change of material quality by analyzing the gradient matrix information of the predicted quality index QTY(t), so that the system can accurately capture the quality fluctuations in the production process. When the material quality status is unqualified, the system automatically iterates the previous quality prediction, production regulation and evaluation steps to achieve closed-loop control and continuous optimization. Through in-depth analysis and real-time feedback of quality trends, not only can timely corrections be made, but also production parameters can be gradually optimized to ensure the stability and high-quality output of the production line during long-term operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 The present invention is a schematic diagram of the steps of a refractory production line control method based on artificial intelligence. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0067] Example 1

[0068] The present invention provides a refractory production line control method based on artificial intelligence, please refer to Figure 1 , including the following steps:

[0069] S1. Several sensors are set up in the production line operation area to collect production line operation status information and environmental status information to form a collection data group PRM, and the collection data group PRM is transmitted in real time through a wireless network and Ethernet;

[0070] S2, performing data preprocessing and data feature extraction on the received collected data group PRM, including denoising preprocessing, filtering preprocessing and principal component analysis, to form a feature data set FTS;

[0071] S3. Establish a quality prediction learning model by using deep convolutional neural network technology, and input the acquired feature data set FTS as the input data set into the quality prediction learning model for training and analysis, obtain the predicted quality index QTY(t), and match it with the preset material production quality status threshold CZ to obtain the material quality status result;

[0072] S4. Triggering production control decision generation according to the material quality status result, obtaining the production control data group CTL(t) at time t, and regulating the production line generation control parameters according to the production control data group CTL(t);

[0073] S5. Evaluate the status of the control result of the production line output materials after regulation at a fixed period, obtain the real-time evaluation coefficient EXE(t) of the production line control at time t, match it with the preset post-regulation evaluation threshold TZ, and obtain the quality status of the material output after the production line regulation for iterative feedback.

[0074] In this embodiment, through steps S1 to S5, multiple sensors are deployed in the production line operation area to collect multi-dimensional production status and environmental information including molding pressure, temperature, running speed and humidity in real time to form a data group PRM, and after data preprocessing steps such as denoising, filtering and principal component analysis, a high-quality feature data set FTS is generated. The deep convolutional neural network technology is used to train and analyze FTS to generate a predicted quality index QTY(t), and the material quality state is accurately judged by matching with the material production quality state threshold CZ, thereby triggering the corresponding production control decision CTL(t). Through the fixed-cycle evaluation and real-time feedback mechanism, the real-time evaluation coefficient EXE(t) of the production line control at time t is obtained, and the production parameters are dynamically adjusted according to the evaluation results to form a closed-loop control system. The integrated solution of multi-dimensional data acquisition and processing, multi-level intelligent analysis and dynamic feedback not only overcomes the limitations of single parameter detection and simple comparison of traditional methods, but also can more comprehensively and accurately control each link of the production process, and also realizes the continuous optimization of production quality through cyclic iteration logic, significantly improving the stability and product quality of the refractory production line.

[0075] Example 2

[0076] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: S1 also includes S11 and S12;

[0077] S11. Several sensors are set in the production line operation area to collect production line operation status information and environmental status information, including molding pressure information, equipment temperature information, operation speed information, environmental humidity information and equipment vibration frequency information, to form a time series data group, including molding pressure Pf(t), equipment temperature Te(t), equipment operation speed Vr(t), environmental humidity He(t) and equipment vibration Vu(t) at time t, which are integrated to form a collection data group PRM at time t, specifically PRM={Pf(t), Te(t), Vr(t), He(t), Vu(t)};

[0078] Among them, the production line operation area includes the raw material input area, molding area, heating area and cooling area; several sensors include temperature sensors, pressure sensors, humidity sensors and vibration sensors;

[0079] S12, the collected data group PRM at time t is transmitted to the data processing in real time by using the wireless network and Ethernet, and is classified and stored according to the data types included in the collected data group PRM.

[0080] Step S2 also includes S21 and S22;

[0081] S21, performing artificial intelligence data preprocessing on the received collected data group PRM, including denoising preprocessing and filtering preprocessing, wherein the denoising preprocessing includes using wavelet transform and mean filtering to perform denoising preprocessing, thereby eliminating high-frequency noise and random fluctuations in the collected data group PRM; the filtering preprocessing includes using a low-pass filter and a Kalman filter to perform filtering preprocessing, thereby eliminating related high-frequency noise in the collected data group PRM after the denoising preprocessing;

[0082] S22. Use principal component analysis to extract artificial intelligence information features from the preprocessed collected data group PRM to obtain the feature PCA1 of molding pressure, the feature PCA2 of equipment temperature, the feature PCA3 of equipment running speed, the feature PCA4 of ambient humidity and the feature PCA5 of equipment vibration to form a feature data set FTS, specifically FTS = {PCA1, PCA2, PCA3, PCA4, PCA5}.

[0083] The feature PCA1 of the molding pressure is obtained by the following feature extraction method:

[0084] PCA1(t)=max(Pf(t))-min(Pf(t));

[0085] Wherein, PCA1(t) represents the characteristic value of the molding pressure at time t, specifically represents the fluctuation amplitude of the molding pressure, max(Pf(t)) represents the peak molding pressure at time t, and min(Pf(t)) represents the valley molding pressure at time t;

[0086] The PCA2 feature of the device temperature is obtained by the following feature extraction method:

[0087]

[0088] Where PCA2(t) represents the device temperature characteristic value at time t, n represents the total number of samples in a fixed period, and Te(t, i) represents the device temperature Te(t) at time t of sample i;

[0089] The PCA3 feature of the equipment running speed is obtained through the following feature extraction method:

[0090]

[0091] Where PCA3(t) represents the characteristic value of the equipment running speed at time t, Vr(t, i) represents the equipment speed at time t of sample i, and Indicates the average value of the device's operating speed;

[0092] The feature PCA4 of ambient humidity is obtained by the following feature extraction method:

[0093] PCA4(t)=min(He(t));

[0094] Wherein, PCA4(t) represents the ambient humidity He(t) at time t, and min(He(t)) represents the valley value of the ambient humidity He(t) at time t;

[0095] The PCA5 feature of the equipment vibration is obtained through the following feature extraction method:

[0096] PCA5(t)=max(Vu(t));

[0097] Where PCA5(t) represents the equipment vibration characteristic value at time t, and max(Vu(t)) represents the peak value of the equipment vibration Vu(t) at time t.

[0098] Example 3

[0099] This embodiment is explained in Example 2. Please refer to Figure 1 , specifically: step S3 also includes S31 and S32;

[0100] S31. Establish a quality prediction learning model by using deep convolutional neural network technology, including multiple convolutional layers, pooling layers and fully connected layers, and input the acquired feature data set FTS as an input data set into the quality prediction learning model to perform artificial intelligence regulation of the quality prediction learning model training process, including using back propagation algorithm and optimization algorithm for regulation, and obtaining the predicted quality index QTY(t) through the analysis output of the output layer of the quality prediction learning model;

[0101] S32, matching the obtained predicted quality index QTY(t) with the preset material production quality status threshold CZ to obtain the material quality status result;

[0102] The material quality status results are obtained by matching:

[0103]

[0104] When the return signal is 1, the material quality status result is qualified;

[0105] When the return signal is 0, the material quality status result is unqualified.

[0106] The predicted quality index QTY(t) is obtained by the following calculation formula:

[0107] QTY(t)=fout(Wfc·fpool(Wconu·FTS(t)+bconu)+bfc);

[0108] Wherein, fout represents the output function, specifically represents the conversion of the output of the fully connected layer into the prediction quality index QTY(t), Wfc represents the weight matrix of the fully connected layer, fpool represents the pooling function, including the use of maximum pooling and average pooling to adjust the calculation amount of the model, Wconu represents the weight matrix of the convolutional layer, bconu represents the bias term of the convolutional layer, bfc represents the bias term of the fully connected layer, Wconu FTS(t)+bconu represents the calculation formula of the convolutional layer, specifically represents the extraction of local features from the input feature data FTS(t), and the adjustment of the expression of local features by adding the bias term bconu of the convolutional layer.

[0109] Step S4 also includes S41 and S42;

[0110] S41, triggering production control decision generation according to the material quality status result, and obtaining the production control data group CTL(t) at time t;

[0111] Among them, the generation of production control decisions is triggered specifically through the following mechanisms;

[0112] When the return signal of the material quality status result is 1, the material quality status result is qualified and does not trigger the generation of production control decision;

[0113] When the return signal of the material quality status result is 0, the material quality status result is in an unqualified state, the production control decision is triggered, and the molding pressure Pf(t), the equipment temperature Te(t) and the equipment running speed Vr(t) are processed, including proportionally adjusting the molding pressure Pf(t), the equipment temperature Te(t) and the equipment running speed Vr(t) in the form of percentage, for example, increasing or decreasing the molding pressure Pf(t), the equipment temperature Te(t) and the equipment running speed Vr(t) by 3%, to obtain the production control data group CTL(t) at time t;

[0114] S42. Generate control parameters for the production line according to the production control data group CTL(t). The control content includes the adjustment of molding pressure, the control of equipment temperature and the adjustment of production line running speed. The control content is specifically executed through PLC and DCS industrial control systems.

[0115] The production control data group CTL(t) is generated by the following calculation formula:

[0116] CTL(t)={Pexec(t), Texec(t), Vexec(t)};

[0117] Wherein, Pexec(t) represents the molding pressure value after the production control is adjusted at time t, Texec(t) represents the equipment temperature value after the production control is adjusted at time t, and Vexec(t) represents the equipment operating speed value after the production control is adjusted at time t.

[0118] In this embodiment, by introducing multiple convolutional layers, pooling layers and fully connected layers in the deep convolutional neural network model, the feature data set FTS is efficiently processed, and the back propagation algorithm and optimization algorithm are used to continuously adjust the model to accurately generate the predicted quality index QTY(t). In addition, the scheme designs a matching mechanism based on QTY(t) and the preset material production quality state threshold CZ to ensure real-time monitoring and judgment of material quality. When the quality state is unqualified, the system automatically triggers the production control decision to make refined adjustments to the molding pressure, equipment temperature and operating speed. Through this mechanism of dynamically adjusting production parameters in percentage form, the scheme can flexibly respond to fluctuations in the production process, ensure the stability of the production process and the continuous optimization of material quality. At the same time, through the PLC and DCS industrial control systems to execute the control content, the scheme realizes the intelligent management and automatic control of the production line, greatly improving the production efficiency and quality consistency. Compared with other methods, this scheme not only focuses on real-time and accuracy, but also quickly corrects deviations through feedback mechanisms, adapts to complex and changing production environments, and provides a more intelligent and adaptive control method for the production line.

[0119] Example 4

[0120] This embodiment is explained in Example 3, please refer to Figure 1 , specifically: step S5 also includes S51 and S52;

[0121] S51, the regulated production line output material is evaluated at a fixed period and the regulation result status is obtained to obtain the production line regulation real-time evaluation coefficient EXE(t) at time t;

[0122] The real-time evaluation coefficient EXE(t) of production line control is obtained by the following calculation formula:

[0123] EXE(t)=δ·CTL(t)+ε·▽QTY(t);

[0124] In the formula, δ represents the control gain coefficient, ε represents the adjustment coefficient, and ▽QTY(t) represents the gradient matrix information of the predicted quality index QTY(t), which is specifically used to represent the trend of the predicted quality index QTY(t) within a fixed period;

[0125] S52. Match the preset post-control evaluation threshold TZ with the production line control real-time evaluation coefficient EXE(t) to obtain the material output quality status after the production line control for iterative feedback.

[0126] The material output quality status after production line regulation is obtained through the following matching methods:

[0127] When the real-time evaluation coefficient EXE(t) ≥ the post-regulation evaluation threshold TZ, the qualified evaluation result of the material output status after the production line regulation is obtained;

[0128] When the real-time evaluation coefficient EXE(t) is less than the post-regulation evaluation threshold TZ, the unqualified evaluation result of the material output status after the production line regulation is obtained, and steps S3, S4 and S5 are iteratively executed.

[0129] In this embodiment, by evaluating the production line after regulation at fixed periods, a real-time evaluation coefficient EXE(t) for production line regulation is generated, and the coefficient is matched with the preset post-regulation evaluation threshold TZ to obtain the quality status of the material after production line regulation in real time. In particular, the real-time evaluation coefficient EXE(t)(t) for production line regulation reflects the trend and change of material quality by analyzing the gradient matrix information of the predicted quality index QTY(t), so that the system can accurately capture the quality fluctuations in the production process. When the material quality status is unqualified, the system automatically iterates and executes the previous quality prediction, production regulation and evaluation steps to achieve closed-loop control and continuous optimization. Through in-depth analysis and real-time feedback of quality trends, not only can timely corrections be made, but also production parameters can be gradually optimized to ensure the stability and high-quality output of the production line during long-term operation, significantly improving the overall intelligent level of the production line.

[0130] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A refractory production line control method based on artificial intelligence, characterized in that: The following steps are involved: S1. Several sensors are set up in the production line operation area to collect production line operation status information and environmental status information to form a collection data group PRM, and the collection data group PRM is transmitted in real time through a wireless network and Ethernet; S2, performing data preprocessing and data feature extraction on the received collected data group PRM, including denoising preprocessing, filtering preprocessing and principal component analysis, to form a feature data set FTS; S3. Establish a quality prediction learning model by using deep convolutional neural network technology, and input the acquired feature data set FTS as the input data set into the quality prediction learning model for training and analysis, obtain the predicted quality index QTY(t), and match it with the preset material production quality status threshold CZ to obtain the material quality status result; S4. Triggering production control decision generation according to the material quality status result, obtaining the production control data group CTL(t) at time t, and regulating the production line generation control parameters according to the production control data group CTL(t); S5. Evaluate the status of the control result of the production line output materials after regulation at a fixed period, obtain the real-time evaluation coefficient EXE(t) of the production line control at time t, match it with the preset post-regulation evaluation threshold TZ, and obtain the quality status of the material output after the production line regulation for iterative feedback.

2. The artificial intelligence-based refractory production line control method according to claim 1, characterized in that: Among them, S1 also includes S11 and S12; S11. Several sensors are set in the production line operation area to collect production line operation status information and environmental status information, including molding pressure information, equipment temperature information, operation speed information, environmental humidity information and equipment vibration frequency information, to form a time series data group, including molding pressure Pf(t), equipment temperature Te(t), equipment operation speed Vr(t), environmental humidity He(t) and equipment vibration Vu(t) at time t, which are integrated to form a collection data group PRM at time t, specifically PRM={Pf(t), Te(t), Vr(t), He(t), Vu(t)}; Among them, the production line operation area includes the raw material input area, molding area, heating area and cooling area; several sensors include temperature sensors, pressure sensors, humidity sensors and vibration sensors; S12, the collected data group PRM at time t is transmitted to the data processing in real time by using the wireless network and Ethernet, and is classified and stored according to the data types included in the collected data group PRM.

3. The artificial intelligence-based refractory production line control method according to claim 1, characterized in that: Wherein S2 also includes S21 and S22; S21, performing artificial intelligence data preprocessing on the received collected data group PRM, including denoising preprocessing and filtering preprocessing, wherein the denoising preprocessing includes using wavelet transform and mean filtering to perform denoising preprocessing, thereby eliminating high-frequency noise and random fluctuations in the collected data group PRM; the filtering preprocessing includes using a low-pass filter and a Kalman filter to perform filtering preprocessing, thereby eliminating related high-frequency noise in the collected data group PRM after the denoising preprocessing; S22. Use principal component analysis to extract artificial intelligence information features from the preprocessed collected data group PRM to obtain the feature PCA1 of molding pressure, the feature PCA2 of equipment temperature, the feature PCA3 of equipment running speed, the feature PCA4 of ambient humidity and the feature PCA5 of equipment vibration to form a feature data set FTS, specifically FTS = {PCA1, PCA2, PCA3, PCA4, PCA5}.

4. The artificial intelligence-based refractory production line control method according to claim 3, characterized in that: The feature PCA1 of the molding pressure is obtained by the following feature extraction method: PCA1(t)=max(Pf(t))-min(Pf(t)); Wherein, PCA1(t) represents the characteristic value of the molding pressure at time t, specifically represents the fluctuation amplitude of the molding pressure, max(Pf(t)) represents the peak molding pressure at time t, and min(Pf(t)) represents the valley molding pressure at time t; The feature PCA2 of the device temperature is obtained by the following feature extraction method: Where PCA2(t) represents the device temperature characteristic value at time t, n represents the total number of samples in a fixed period, and Te(t, i) represents the device temperature Te(t) at time t of sample i; The feature PCA3 of the device running speed is obtained by the following feature extraction method: Where PCA3(t) represents the characteristic value of the equipment running speed at time t, Vr(t, i) represents the equipment speed at time t of sample i, and Indicates the average value of the device's operating speed; The feature PCA4 of the ambient humidity is obtained by the following feature extraction method: PCA4(t)=min(He(t)); Wherein, PCA4(t) represents the ambient humidity He(t) at time t, and min(He(t)) represents the valley value of the ambient humidity He(t) at time t; The PCA5 feature of the device vibration is obtained by the following feature extraction method: PCA5(t)=max(Vu(t)); Where PCA5(t) represents the equipment vibration characteristic value at time t, and max(Vu(t)) represents the peak value of the equipment vibration Vu(t) at time t.

5. The artificial intelligence-based refractory production line control method according to claim 1, characterized in that: Wherein S3 also includes S31 and S32; S31. Establish a quality prediction learning model by using deep convolutional neural network technology, including multiple convolutional layers, pooling layers and fully connected layers, and input the acquired feature data set FTS as an input data set into the quality prediction learning model to perform artificial intelligence regulation of the quality prediction learning model training process, including using back propagation algorithm and optimization algorithm for regulation, and obtaining the predicted quality index QTY(t) through the analysis output of the output layer of the quality prediction learning model; S32, matching the obtained predicted quality index QTY(t) with the preset material production quality status threshold CZ to obtain the material quality status result; The material quality status results are obtained by the following matching method: When the return signal is 1, the material quality status result is qualified; When the return signal is 0, the material quality status result is unqualified.

6. The artificial intelligence-based refractory production line control method according to claim 5, characterized in that: The predicted quality index QTY(t) is obtained by the following calculation formula: QTY(t)=fout(Wfc·fpool(Wconu·FTS(t)+bconu)+bfc); Wherein, fout represents the output function, specifically represents the conversion of the output of the fully connected layer into the prediction quality index QTY(t), Wfc represents the weight matrix of the fully connected layer, fpool represents the pooling function, including the use of maximum pooling and average pooling to adjust the calculation amount of the model, Wconu represents the weight matrix of the convolutional layer, bconu represents the bias term of the convolutional layer, bfc represents the bias term of the fully connected layer, Wconu FTS(t)+bconu represents the calculation formula of the convolutional layer, specifically represents the extraction of local features from the input feature data FTS(t), and the adjustment of the expression of local features by adding the bias term bconu of the convolutional layer.

7. The artificial intelligence-based refractory production line control method according to claim 1, characterized in that: Among them, S4 also includes S41 and S42; S41, triggering production control decision generation according to the material quality status result, and obtaining the production control data group CTL(t) at time t; Among them, the generation of production control decisions is triggered specifically through the following mechanisms; When the return signal of the material quality status result is 1, the material quality status result is qualified and does not trigger the generation of production control decision; When the return signal of the material quality status result is 0, the material quality status result is unqualified, which triggers the production control decision generation. By processing the molding pressure Pf(t), equipment temperature Te(t) and equipment running speed Vr(t), the production control data group CTL(t) at time t is obtained; S42. Generate control parameters for the production line according to the production control data group CTL(t). The control content includes the adjustment of molding pressure, the control of equipment temperature and the adjustment of production line running speed. The control content is specifically executed through PLC and DCS industrial control systems.

8. The artificial intelligence-based refractory production line control method according to claim 7, characterized in that: The production control data set CTL(t) is generated by the following calculation formula: CTL(t)={Pexec(t), Texec(t), Vexec(t)}; Wherein, Pexec(t) represents the molding pressure value after the production control is adjusted at time t, Texec(t) represents the equipment temperature value after the production control is adjusted at time t, and Vexec(t) represents the equipment operating speed value after the production control is adjusted at time t.

9. The artificial intelligence-based refractory production line control method according to claim 1, characterized in that: Among them, S5 also includes S51 and S52; S51, the regulated production line output material is evaluated at a fixed period and the regulation result status is obtained to obtain the production line regulation real-time evaluation coefficient EXE(t) at time t; The production line control real-time evaluation coefficient EXE(t) is obtained by the following calculation formula: EXE(t)=δ·CTL(t)+ε·▽QTY(t); In the formula, δ represents the control gain coefficient, ε represents the adjustment coefficient, and ▽QTY(t) represents the gradient matrix information of the predicted quality index QTY(t), which is specifically used to represent the trend of the predicted quality index QTY(t) within a fixed period; S52. Match the preset post-control evaluation threshold TZ with the production line control real-time evaluation coefficient EXE(t) to obtain the material output quality status after the production line control for iterative feedback.

10. The artificial intelligence-based refractory production line control method according to claim 9, characterized in that: The material output quality status after the production line is regulated is obtained by the following matching method: When the real-time evaluation coefficient EXE(t) ≥ the post-regulation evaluation threshold TZ, the qualified evaluation result of the material output status after the production line regulation is obtained; When the real-time evaluation coefficient EXE(t) is less than the post-regulation evaluation threshold TZ, the unqualified evaluation result of the material output status after the production line regulation is obtained, and steps S3, S4 and S5 are iteratively executed.

Images