A mold production information data analysis method and system

Abstract

The present invention relates to the field of data processing, and specifically to a mold production information data analysis method and system, which collects temperature data of a forklift axle housing mold heat treatment process in real time, uses an LSTM model to predict the temperature, and calculates the thermal stress index and thermal response coefficient based on the physical and environmental characteristics of the mold, optimizes the adaptive learning rate of the LSTM model, and improves the stability and accuracy of model training. Finally, based on the model prediction, the temperature control system is adjusted through a fuzzy control strategy to achieve intelligent control of the mold temperature and improve the mold production quality and service life. The present invention combines the LSTM model prediction driven by real-time temperature data with the fuzzy control strategy to achieve adaptive learning rate optimization and intelligent temperature control based on the thermal stress response coefficient.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a mold production information data analysis method and system. Background Art

[0002] In the industrial field, the forklift axle housing is one of the important parts of the forklift, which is used to support and transfer the load of the wheels, and cooperate with the axle system to ensure the stable driving and steering performance of the forklift. The forklift axle housing mold is mainly used to cast the metal liquid so that the geometry and size of the axle housing meet the design requirements. In order to ensure the strength, wear resistance and fatigue resistance of the axle housing, the mold needs to have high precision and good durability. Usually, the production process of the forklift axle housing mold includes design, rough machining and finishing, heat treatment and surface treatment. Among them, the heat treatment process is an important part of mold making. The main purpose is to improve the hardness, strength, toughness and wear resistance of the mold to cope with the high-intensity use requirements in long-term production. If the temperature control during the heat treatment process is inaccurate or unstable, it may cause problems such as cracks, dimensional deformation, insufficient hardness or increased brittleness in the mold. Therefore, accurate control of temperature changes during the heat treatment process is the key to ensuring the quality and service life of the forklift axle housing mold.

[0003] In the prior art, some prediction models are usually used to correct and regulate the current actual temperature. For example, the Chinese patent application document with publication number CN118485236A discloses a heat supply system heat station regulation prediction method and system based on XGBoost, which establishes a heat supply station regulation model by collecting relevant data in the heat supply system, and trains the heat supply station regulation model using the XGBoost algorithm; according to the predicted secondary water supply temperature and the delay time of the secondary pipe network temperature response, the heat supply station regulation strategy is adjusted, thereby realizing the accurate prediction and regulation of the secondary supply and return water temperatures of the heat supply station. However, this method has some technical defects: the parameters of its regression model, especially the learning rate, are fixed in the training stage, and the model parameters cannot be fine-tuned in real time according to new data during operation, which makes the model not flexible enough to respond to changes in the environment and system, and is prone to fitting or underfitting, which reduces the generalization ability of the model. Summary of the invention

[0004] In view of the problem that the above-mentioned model is not flexible enough to respond to changes in the environment and the system, in the first aspect, the present invention proposes a mold production information data analysis method and system, including: training an LSTM prediction model based on historical normal temperature data of the mold heat treatment process; calculating the mean of the thermal response coefficients of multiple sample windows contained in each batch of training data; obtaining an adaptive learning rate of the LSTM prediction model, and the adaptive learning rate is positively correlated with the mean and the set initial learning rate; the specific process of obtaining the thermal response coefficient is: collecting the center temperature and the distal temperature of the mold, and the distal temperature is the mean of the temperatures on both sides of the mold; the center temperature and the distal temperature at multiple collection times together constitute a sample window; the historical normal The temperature data includes multiple sample windows; the temperature difference between all the center temperatures and the distal temperatures in each sample window is calculated to obtain a temperature difference sequence; a polynomial fitting is performed on the temperature difference sequence to obtain the corresponding quadratic term coefficient; the product of the quadratic term coefficient, the elastic modulus of the mold material at the average temperature of the corresponding sample window, and the thermal expansion coefficient is used as the thermal stress index of the corresponding sample window; the thermal response coefficient of each sample window is obtained, and the thermal response coefficient is positively correlated with the thermal stress index of the corresponding sample window and the average thickness of the mold, and negatively correlated with the surface area of ​​the mold; the temperature data of the current mold is predicted based on the trained LSTM prediction model to obtain the predicted temperature, and the heating power is regulated based on the temperature error between the current predicted temperature and the actual temperature.

[0005] The present invention proposes a mold production information data analysis method based on LSTM model and adaptive learning rate, which collects temperature data of the mold center and remote end in real time, calculates thermal stress index and thermal response coefficient in combination with material properties, and adjusts the learning speed of LSTM model through adaptive learning rate driven by eigenvalue, so that the model converges quickly in high stress areas and performs fine adjustment when the temperature difference is small. Compared with the traditional model using fixed learning rate, the adaptive learning rate of the present invention can dynamically adjust the training strategy according to the current working conditions, solving the problem of slow response of the model under complex working conditions. Therefore, the present invention ensures the high efficiency and stable training of the model and the accuracy of temperature prediction in complex heat treatment environment, and improves the production quality and service life of the mold.

[0006] Furthermore, the calculation method of the thermal stress index is specifically as follows:

[0007]

[0008] in Represents the sample window Thermal stress index; Indicates that the mold is in the sample window Average temperature inside The elastic modulus under Indicates that the mold is in the sample window Average temperature inside The coefficient of thermal expansion under is the maximum and minimum normalization function.

[0009] Furthermore, the calculation method of the thermal response coefficient is specifically as follows:

[0010]

[0011] in Represents the sample window The corresponding thermal stress index; U represents the average thickness of the mold; S represents the surface area of ​​the mold; represents the normalization function; represents the natural exponential function.

[0012] By combining multiple factors such as thermal stress index and physical characteristics of materials, the mold's response ability to temperature changes is calculated. This response coefficient not only takes into account the internal characteristics of the material, but also combines the influence of the mold structure (such as thickness) to make the model more accurate in predicting temperature changes. Compared with the single temperature monitoring in the prior art, the thermal response coefficient of the present invention provides a more comprehensive system response evaluation by comprehensively considering multiple physical characteristics.

[0013] Furthermore, the calculation method of the modified learning rate is specifically as follows:

[0014]

[0015] in Represents the corrected learning rate of the yth batch of training data; represents the initial learning rate; Represents the mean value of the thermal response coefficients corresponding to all sample windows contained in the yth batch of training data; represents the natural exponential function.

[0016] The adaptive learning rate adjustment method based on the mean value of the thermal response coefficient dynamically adjusts the learning speed of the model according to the thermal response coefficient of each batch during the training process of the LSTM model. This method enables the LSTM model to accelerate convergence when the temperature difference is large and slow down the training pace when the temperature difference is small, thereby achieving adaptive optimization of the model training process. Compared with the fixed learning rate of the traditional model, the present invention dynamically adjusts the learning rate so that the model can respond quickly when the temperature difference and stress change, effectively reducing the overfitting or underfitting problems caused by improper learning pace during the training process.

[0017] Furthermore, the heating power is regulated based on the error between the current predicted temperature and the actual temperature, and further includes: using a fuzzy control algorithm to regulate the heating power of the thermal management system based on a set control rule, wherein the control rule is specifically: when the temperature error is greater than And the error change rate is greater than When the temperature error is within When the temperature error is within And the error change rate is , the heating power remains unchanged; when the temperature error is less than And the error change rate is less than When the temperature error is within Inside.

[0018] By analyzing the temperature error and change rate predicted by the LSTM model, the mold heating and cooling power is dynamically adjusted. The fuzzy control system can flexibly adjust the power according to the changing trend of the predicted results, while the traditional PID control system often needs to adjust the parameters repeatedly to achieve better results when dealing with complex temperature fluctuations. The fuzzy control algorithm of the present invention can adaptively adjust the control strategy according to the changes in real-time temperature data, reducing the time cost of manual parameter adjustment and improving the control accuracy and system response speed.

[0019] Furthermore, collecting the center temperature and remote temperature of the mold also includes: fixing a thermocouple temperature sensor at the center of the mold slot opening to collect the center temperature of the mold; fixing thermocouple temperature sensors at both ends of the mold to collect the temperatures on both sides of the mold, and taking the average of the temperatures on both sides as the remote temperature; setting the sensor collection frequency to once every 30 seconds.

[0020] By arranging temperature sensors at the center and remote ends of the mold, temperature data is collected regularly and used as input for the LSTM model. This method ensures comprehensive monitoring of the internal temperature of the mold and provides reliable data support for prediction and regulation. Compared with the traditional single-point temperature collection method, the present invention more comprehensively reflects the temperature distribution of the mold through multi-point collection, reducing the prediction error caused by local temperature anomalies. At the same time, the data collected at multiple points makes the prediction of the LSTM model more accurate and improves the controllability of the mold heat treatment process.

[0021] Furthermore, the training process of the LSTM prediction model also includes: using the first 70% of the historical normal temperature data of the mold heat treatment process as a training set; using the middle 15% of the historical normal temperature data as a validation set; and using the last 15% of the historical normal temperature data as a test set.

[0022] In a second aspect, the present invention provides a mold production information data analysis system, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the mold production information data analysis method of the present invention is implemented.

[0023] The technical effects of the present invention are:

[0024] The present invention predicts the future temperature change of the mold through the LSTM model, and uses the fusion characteristic value based on temperature difference and thermal stress to adjust the learning rate, thus realizing the intelligent prediction and regulation. This method ensures that the prediction model can respond quickly when dealing with different working conditions, and significantly improves the production quality of forklift axle housing molds. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0026] Figure 1 is a flowchart schematically showing a mold production information data analysis method in an embodiment of the present invention;

[0027] Figure 2 is a temperature timing diagram schematically showing a heat treatment process of a forklift axle housing mold in an embodiment of the present invention;

[0028] Figure 3 is a schematic diagram schematically showing the placement of sensors in an embodiment of the present invention;

[0029] Figure 4 It is a block diagram schematically showing the structure of the mold production information data analysis system in an embodiment 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 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0031] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0032] Example of mold production information data analysis method:

[0033] like Figure 1 As shown, the mold production information data analysis method of the present invention includes:

[0034] S1. Obtain the historical temperature data of the heat treatment process of the forklift axle housing mold.

[0035] The forklift axle housing is a key component of the forklift axle system, responsible for carrying the load of the wheels and cooperating with the axle system to ensure the stable driving and steering functions of the forklift. The forklift axle housing mold is an important tool for realizing the mass production of axle housings. The mold is usually made of high-strength steel and undergoes complex processing and heat treatment processes to ensure that it can meet the needs of producing high-strength and wear-resistant axle housings.

[0036] Heat treatment is a very important step in the production of forklift axle housing molds, which usually includes three steps: preheating, quenching and tempering. In one embodiment, P20 steel can be used as the production material of the forklift axle housing mold, such as Figure 2 As shown, it shows the overall temperature change process of the forklift bridge housing mold during heat treatment over time. Stage A in the figure represents the preheating process of the mold. The function of this stage is to slowly heat up the mold to prevent the mold from generating excessive thermal stress due to sudden temperature changes, causing cracks or deformation, and to prepare for subsequent quenching. Stage B in the figure is the quenching process of the mold, which is also the most important stage in the entire heat treatment process. It heats the steel to the austenite zone and then quickly cools the steel to transform austenite into martensite, thereby improving the hardness and strength of the mold. In this embodiment, oil extraction is used to complete the cooling process. Stage C in the figure represents the tempering process of the mold, the purpose of which is to eliminate the internal stress generated after quenching, prevent the mold from cracking or deformation, reduce the brittleness of the mold and improve the toughness of the mold.

[0037] In summary, in the production of forklift axle housing molds, the heat treatment process is a process in which the temperature changes frequently, and the temperature needs to be precisely controlled, otherwise it may cause defects of different degrees or types in the mold. Since the forklift axle housing mold is relatively large as a whole, its own thermal response may be slow when the temperature of the heating system changes. Therefore, in this embodiment, a prediction model can be used to predict the temperature based on the real-time temperature of the mold, obtain the future temperature change of the mold, and control the heating system based on this change to ensure that the temperature change of the mold itself is within a normal range, thereby ensuring the quality of the forklift axle housing mold.

[0038] In order to facilitate the production of forklift axle housings, the forklift axle housing mold itself is a rectangular body with a slot-shaped opening, and the mold itself is relatively thick. This structural characteristic leads to increased heat convection in the slot and poor heat dissipation during heating, and its temperature will be correspondingly higher than the temperature on both sides of the mold. Therefore, in this embodiment, based on its structural characteristics, thermocouple temperature sensors can be fixed at the center of the mold slot opening and at both ends of the mold to collect temperature changes in different parts of the mold during temperature changes. Since the mold has a large thermal response, the sensor collection frequency can be set to once every 30 seconds. Figure 3 As shown, A sensor that indicates the center of the mold slot opening. and Represents the sensors on both sides of the mold, that is, for any acquisition time , three temperature parameters will be obtained , as well as .because and Both are the temperatures at both ends of the mold, and the structural characteristics of the sensor locations are the same, so during the heating process of the mold and There will not be a large deviation, so when analyzing the temperature of the mold later, just take the average of these two temperatures. The distal temperature and center temperature of the lower mold are respectively and , and there are:

[0039]

[0040] After obtaining the temperature of the sensor on the mold each time, it is uploaded to the database synchronously. At the same time, in order to ensure the integrity and accuracy of the data in subsequent analysis, the data stored in the database needs to be cleaned, including outlier detection, missing value processing and data normalization. In this embodiment, you can use The outlier detection is performed according to the principle, and then the missing value is processed by linear interpolation, and finally the maximum and minimum normalization is performed. The above data cleaning algorithms belong to the known technology, and the specific implementation methods are not described here. It should be noted that the molds mentioned later in this embodiment refer to the forklift axle housing molds, and the subsequent parts will not be described.

[0041] By arranging temperature sensors at the center and remote ends of the mold, temperature data is collected regularly and used as input for the LSTM model. This method ensures comprehensive monitoring of the internal temperature of the mold and provides reliable data support for prediction and regulation. Compared with the traditional single-point temperature collection method, the present invention more comprehensively reflects the temperature distribution of the mold through multi-point collection, reducing the prediction error caused by local temperature anomalies. At the same time, the data collected at multiple points makes the prediction of the LSTM model more accurate and improves the controllability of the mold heat treatment process.

[0042] S2. Build an LSTM model based on the production information of the heat treatment process of forklift axle mold.

[0043] LSTM is a recurrent neural network (RNN) suitable for time series data, which can capture time dependency and long-term and short-term correlation characteristics. In this embodiment, the temperature of the forklift axle housing mold heat treatment process has obvious time dependency, so the LSTM model is more suitable for predicting the temperature change of the mold in the next several time steps.

[0044] In this embodiment, the training data of the LSTM model is obtained from the database, specifically: All correspond and , then construct a sample window with a time step of 10 as the empirical value , as shown below:

[0045]

[0046] Each row represents the temperature data of one time step; each column represents the remote temperature and the central temperature at different acquisition times; in this embodiment, in addition to selecting the empirical value 10 as the time step, the implementer can also select other step values ​​such as 15, 20 or 25 according to actual conditions, that is, in this embodiment, the temperature data of 10 time steps are used to predict the temperature data of the next time step.

[0047] Obtain a temperature time series data set during a normal and complete mold production heat treatment process, and use a sliding window with a sliding step of 1 to construct the temperature time series data set as a sample data set. For example, there is a two-dimensional temperature time series for:

[0048]

[0049] in Indicates the center temperature at the first acquisition moment With remote temperature ; Indicates the center temperature at the last acquisition time h With remote temperature . Then the first sample window and the second sample window constructed are:

[0050]

[0051]

[0052] Slide the window sequentially to build multiple sample windows until the last sample window contains the two-dimensional temperature parameters , complete the construction of the sample data set. In this embodiment, the first 70% of the data in the sample data set is used as a training set, the middle 15% of the data is used as a validation set, and the last 15% of the data is used as a test set. The network structure of LSTM includes an output layer, a hidden layer, a fully connected layer, and an output layer, wherein the training set is input into the input layer; two hidden layers are set and the number of units in each hidden layer is set to an empirical value of 50; the output mapping of the fully connected layer is set to the temperature prediction value for the next time step; the number of units in the output layer is set to two, and the predicted center temperature and the predicted remote temperature can be output simultaneously. In the LSTM model training process of this embodiment, the mean square error can also be used as a loss function. The rest of the LSTM algorithm belongs to the known technology and will not be repeated here.

[0053] In summary, the LSTM model is trained using a normal and complete temperature time series data set in the mold production heat treatment process in the database to obtain a trained temperature prediction model. The temperature prediction model can be used to predict future temperature data using the current real-time temperature data collected, and then the heating system can be adjusted to keep the mold temperature within the set range.

[0054] S3. Build a sample window and obtain a thermal stress index. Based on the thermal stress index, obtain the thermal response coefficient of the forklift axle housing mold during heat treatment and correct the learning rate of the LSTM model training.

[0055] In the original LSTM model obtained in step S2, the learning rate is usually a fixed value, so problems may occur in the model during training, such as insufficient weight update speed when the temperature difference is large, and the error cannot be corrected in time; and when the temperature is stable, the too fast learning rate causes the model to oscillate and is difficult to converge. In this case, the Adam optimizer is generally used to regulate the learning rate, which is suitable for most deep learning tasks, but it is only adjusted based on the statistical characteristics of the gradient and cannot consider the physical and environmental characteristics in this scenario. Therefore, in this embodiment, the learning rate in the LSTM model is optimized according to the mold characteristics.

[0056] Thermal stress is stress caused by uneven temperature distribution inside the material. For example, when the temperature in the center of the mold rises, while the temperature on both sides (the temperature at the far end) is relatively low, the temperature difference between the two may cause inconsistent expansion and contraction of the material, thereby generating thermal stress. If this thermal stress exceeds the tensile strength or fatigue limit of the mold material, the mold will crack or permanently deform, resulting in quality problems. During the heat treatment process of the mold, the thermal stress inside the mold cannot be directly detected, but because the change in thermal stress is closely related to the change in mold temperature, in this embodiment, the change in thermal stress inside the mold can be reflected as much as possible by analyzing the change in mold temperature.

[0057] S301, construct a sample window and obtain a thermal stress index.

[0058] In step S2, multiple sample windows are constructed. For any sample window have:

[0059]

[0060] It represents the mold center temperature and distal temperature for 10 consecutive time steps, where i represents the i-th time step in the sample window. For the heat treatment process of the mold, the temperature difference between the mold center temperature and the distal temperature reflects the inconsistency of thermal expansion between different parts and is the direct source of stress. Therefore, the temperature difference of each time step in each sample window is first obtained, as shown below:

[0061]

[0062] The first two columns in the above matrix are the initial data of the sample window, and the third column is the temperature difference between the center temperature and the far end temperature of the mold at each time step, which is the temperature difference sequence of this sample window. , obtain its temperature difference series and record it as Here, the polynomial regression algorithm is used to fit the temperature difference series, and the quadratic term coefficient is used to determine the trend change of the temperature difference series. For any sample window , there are quadratic coefficients The above polynomial regression algorithm is a well-known technology and will not be described in detail here.

[0063] In any sample window The quadratic coefficient of After that, calculate any sample window The average temperature in , the specific calculation method is:

[0064]

[0065] So far, any sample window is obtained The quadratic coefficient of and average temperature , calculate the thermal stress index of the mold in the heat treatment stage represented by this sample window, and record it as , the specific calculation method is:

[0066]

[0067] in Indicates that the mold is in the sample window Average temperature inside The elastic modulus under the condition of 400°C is shown in the figure below: for ; Indicates that the mold is in the sample window Average temperature inside The thermal expansion coefficient under the following conditions is exemplified: In this embodiment, P20 steel is used, and its thermal expansion coefficient at about 400°C in the preheating stage is for ; Represents the temperature difference series The coefficient of the quadratic term after polynomial regression fitting; It is a maximum and minimum normalization function, which is used to standardize the quadratic term coefficients after polynomial regression fitting of all sample windows to facilitate the subsequent analysis of its quantitative characteristics.

[0068] The elastic modulus and thermal expansion coefficient Together they determine the mold material's response speed to temperature differences and the degree of stress accumulation under the current temperature conditions. The larger the value, the larger the sample window. The change of internal temperature difference has an upward trend, that is, the sample window The temperature difference gradually expands, indicating that in the sample window During the heat treatment stage, the temperature difference between the center and the distal end of the mold is unstable and tends to expand. At this time, the greater the thermal stress inside the mold, the greater the sample window. Corresponding thermal stress index It should be noted that the thermal stress index here does not represent the accurate thermal stress value. Since the thermal stress value cannot be accurately calculated through finite element analysis and other methods during the mold production process, the thermal stress index is calculated here to reflect the change of thermal stress.

[0069] S302, obtaining a thermal response coefficient during heat treatment of a forklift axle housing mold based on a thermal stress index.

[0070] The above analysis mainly analyzes the changes in thermal stress that may be generated during the heat treatment process of the mold during production, and further considers the heat dissipation performance of the mold to obtain any sample window The thermal response coefficient is denoted as , the specific calculation process is:

[0071]

[0072] in Represents the sample window The corresponding thermal stress index; Indicates the average thickness of the mold; Indicates the surface area of ​​the mold; represents the normalization function, which is used to normalize the thermal response coefficient to the maximum and minimum; It represents the natural exponential function, which reflects the buffering effect of the material on the thermal response when the heat dissipation performance increases to a certain extent.

[0073] when The larger the value, the larger the sample window. The more likely the thermal stress inside the mold is to gradually increase during the corresponding heat treatment stage, the higher the thermal response coefficient corresponding to the sample window. The larger the average thickness U of the mold, the slower the heat exchange between the mold and the outside world, which is more likely to cause heat accumulation inside the mold, the lower the heat dissipation efficiency of the mold, and the increase in the internal temperature gradient. The corresponding sample window The corresponding thermal response coefficient When S is larger, the area of ​​contact between the mold and the cooling medium or air is larger, the effective heat dissipation area is larger, the heat dissipation efficiency of the mold is greater, and the temperature gradient inside the mold will decrease, thereby reducing the risk of thermal stress accumulation. The corresponding sample window The corresponding thermal response coefficient The smaller it is.

[0074] By combining multiple factors such as thermal stress index and physical characteristics of materials, the mold's response ability to temperature changes is calculated. This response coefficient not only takes into account the internal characteristics of the material, but also combines the influence of the mold structure (such as thickness) to make the model more accurate in predicting temperature changes. Compared with the single temperature monitoring in the prior art, the thermal response coefficient of the present invention provides a more comprehensive system response evaluation by comprehensively considering multiple physical characteristics.

[0075] S303: Obtain an adaptive learning rate based on the thermal response coefficient.

[0076] In summary, for any sample window , all have corresponding thermal response coefficients , which is used to quantify the comprehensive response ability of the mold to temperature changes during the heat treatment process. The larger it is, the greater the risk of thermal stress concentration in the mold. Therefore, in this embodiment, the learning rate of model training is adaptively adjusted based on the thermal response coefficient of each sample window. The specific method is:

[0077] During the LSTM model training process, the number of samples processed in each training is generally 32 or 64. In this embodiment, the number of samples is selected as the empirical value of 32, and the number of training rounds is selected as the empirical value of 50. In each batch of training, there is a corresponding learning rate for adjusting the step size of the weight. In this embodiment, the initial learning rate can be set to , and there is , and then calculate the mean thermal response coefficient of all sample windows in each batch to adaptively adjust the learning rate during the training of the batch. Exemplary explanation: For the yth batch of training sample data sets, which contains 32 consecutive sample windows, and each sample window has a corresponding thermal response coefficient, then calculate the mean thermal response coefficient of these 32 sample windows and record it as , then the learning rate of the y-th batch of training sample data sets is:

[0078]

[0079] in Represents the learning rate of the yth batch of training sample data sets; represents the initial learning rate; Represents the mean value of the thermal response coefficients corresponding to all sample windows contained in the y-th batch of training sample data sets; represents the natural exponential function.

[0080] when The larger the value, the larger the thermal response coefficient of the continuous sample windows in the yth batch. The mold has a greater risk of thermal stress concentration in the heat treatment stage corresponding to these sample windows. The faster the weight adjustment speed should be, the faster the prediction error accumulation should be avoided. The learning rate of the yth batch training sample data set is The larger the value, the exponential function is used in this embodiment to reflect the rapid response of the learning rate to the large mean value of the thermal response coefficient; when The smaller it is, the smaller the thermal response coefficient of the continuous sample windows in the yth batch is. The risk of thermal stress concentration in the heat treatment stage corresponding to these sample windows is low. The adjustment speed of the weights should be slowed down to ensure that the model is finely adjusted in the stable area to avoid oscillation. The learning rate of the yth batch training sample data set is The smaller.

[0081] The adaptive learning rate adjustment method based on the mean value of the thermal response coefficient dynamically adjusts the learning speed of the model according to the thermal response coefficient of each batch during the training process of the LSTM model. This method enables the LSTM model to accelerate convergence when the temperature difference is large and slow down the training pace when the temperature difference is small, thereby achieving adaptive optimization of the model training process. Compared with the fixed learning rate of the traditional model, the present invention dynamically adjusts the learning rate so that the model can respond quickly when the temperature difference and stress change, effectively reducing the overfitting or underfitting problems caused by improper learning pace during the training process.

[0082] At this point, in the original LSTM model training process, the learning rate of each batch of training data is dynamically adjusted in combination with the thermal response coefficient of the sample window, the LSTM model with improved adaptive learning rate is obtained and the entire training process of the model is completed.

[0083] S4. Predict the temperature based on the improved LSTM model and use fuzzy control to regulate the current thermal management system.

[0084] In step S3, the improved trained LSTM model is obtained, and the center temperature and distal temperature of the current mold in the heat treatment process for a total of 10 time steps are collected, and the center temperature and distal temperature of the next time step are predicted in the trained LSTM model. The center temperature and distal temperature of the next time step are compared with the actual center temperature and distal temperature. The temperature control system is regulated by the fuzzy control algorithm to keep the mold within a reasonable safety temperature range during production, thereby ensuring the quality of the finished mold. The specific fuzzy control strategy is as follows:

[0085] The temperature error is set, that is, the difference between the predicted temperature and the actual temperature. At the same time, since the center temperature and the remote temperature are collected at the same time in this embodiment, the temperature error is calculated as follows: first, the center predicted temperature and the remote predicted temperature of the 11th time step based on the current 10 time steps are obtained, and then the center actual temperature and the remote actual temperature of the current 11th time step are obtained. The temperature difference between the center predicted temperature and the center actual temperature is used as the first temperature difference, and the temperature difference between the remote predicted temperature and the remote actual temperature is used as the second temperature difference. The first temperature difference and the second temperature difference are calculated to obtain the temperature error.

[0086] In this embodiment, the input of fuzzy control includes: setting When the temperature error is low, When the temperature error is greater than When the temperature error rate of change is less than When the temperature error change rate is When the temperature error change rate is greater than When , the temperature error change rate increases. The fuzzy rule base is set later, as follows:

[0087] Rule 1: If the temperature error is high and the error change rate increases, then the system control power is high; Rule 2: If the temperature error is medium and the error change rate is stable, then the system control power remains unchanged; Rule 3: If the temperature error is low and the error change rate decreases, then the system control power is low; and in this embodiment, the centroid method is used to convert the fuzzy result into a specific heating power adjustment value. The above fuzzy control algorithm belongs to the known technology, and the rest will not be described here.

[0088] By analyzing the temperature error and change rate predicted by the LSTM model, the mold heating and cooling power is dynamically adjusted. The fuzzy control system can flexibly adjust the power according to the changing trend of the predicted results, while the traditional PID control system often needs to adjust the parameters repeatedly to achieve better results when dealing with complex temperature fluctuations. The fuzzy control algorithm of the present invention can adaptively adjust the control strategy according to the changes in real-time temperature data, reducing the time cost of manual parameter adjustment and improving the control accuracy and system response speed.

[0089] In summary, the error between the current temperature predicted by the improved LSTM model and the actual temperature is obtained, and the temperature control of the forklift axle mold during the heat treatment process is completed based on the set fuzzy control strategy according to the temperature error, thereby ensuring the production quality of the mold.

[0090] The present invention proposes a mold production information data analysis method based on LSTM model and adaptive learning rate, which collects temperature data of the mold center and remote end in real time, calculates thermal stress index and thermal response coefficient in combination with material properties, and adjusts the learning speed of LSTM model through adaptive learning rate driven by eigenvalue, so that the model converges quickly in high stress areas and performs fine adjustment when the temperature difference is small. Compared with the traditional model using fixed learning rate, the adaptive learning rate of the present invention can dynamically adjust the training strategy according to the current working conditions, solving the problem of slow response of the model under complex working conditions. Therefore, the present invention ensures the high efficiency and stable training of the model and the accuracy of temperature prediction in complex heat treatment environment, and improves the production quality and service life of the mold.

[0091] Example of mold production information data analysis system:

[0092] On the other hand, the present invention also provides a mold production information data analysis system. Figure 4 As shown, the mold production information data analysis system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a mold production information data analysis method according to the first aspect of the present invention is implemented.

[0093] The mold production information data analysis system also includes other components familiar to those skilled in the art, such as a communication interface, and its configuration and functions are known in the art, so they will not be described in detail here.

[0094] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0095] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0096] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A mold production information data analysis method, characterized in that: Methods include: Train the LSTM prediction model based on the historical normal temperature data of the mold heat treatment process; Calculate the mean value of the thermal response coefficients of multiple sample windows contained in each batch of training data; Obtain the adaptive learning rate of the LSTM prediction model. The adaptive learning rate is positively correlated with the mean and the set initial learning rate. The specific process of obtaining the thermal response coefficient is as follows: The center temperature and the distal temperature of the mold are collected, and the distal temperature is the average of the temperatures on both sides of the mold; the center temperature and the distal temperature at multiple collection times together constitute a sample window; the historical normal temperature data contains multiple sample windows; the temperature difference between all the center temperatures and the distal temperatures in each sample window is calculated to obtain a temperature difference sequence; a polynomial fitting is performed on the temperature difference sequence to obtain the corresponding quadratic term coefficient; the product of the quadratic term coefficient, the elastic modulus of the mold material at the average temperature of the corresponding sample window, and the thermal expansion coefficient is used as the thermal stress index of the corresponding sample window; The thermal response coefficient of each sample window is obtained. The thermal response coefficient is positively correlated with the thermal stress index of the corresponding sample window and the average thickness of the mold, and negatively correlated with the surface area of ​​the mold. Thermal stress index calculation method: ； in Represents the sample window Thermal stress index; Indicates that the mold is in the sample window Average temperature inside The elastic modulus under Indicates that the mold is in the sample window Average temperature inside The coefficient of thermal expansion under is the maximum and minimum normalization function; Thermal response coefficient calculation method: ； Where U represents the average thickness of the mold; S represents the surface area of ​​the mold; represents the natural exponential function; Based on the trained LSTM prediction model, the temperature data of the current mold is predicted to obtain the predicted temperature. Based on the temperature error between the current predicted temperature and the actual temperature, the heating power is regulated. The heating power of the thermal management system is regulated using a fuzzy control algorithm based on the set control rules. The specific control rules are: When the temperature error is greater than And the error change rate is greater than When the temperature error is within Inside; When the temperature error is And the error change rate is , the heating power remains unchanged; When the temperature error is less than And the error change rate is less than When the temperature error is within Inside.

2. A mold production information data analysis method according to claim 1, characterized in that: The calculation method of the corrected learning rate is as follows: ； in Represents the corrected learning rate of the yth batch of training data; represents the initial learning rate; Represents the mean value of the thermal response coefficients corresponding to all sample windows contained in the yth batch of training data; represents the natural exponential function.

3. A mold production information data analysis method according to claim 1, characterized in that: Collect the center temperature and remote temperature of the mold, including: A thermocouple temperature sensor is fixed at the center of the mold slot opening to collect the center temperature of the mold; Thermocouple temperature sensors are fixed at both ends of the mold to collect the temperatures on both sides of the mold, the average of the temperatures on both sides is used as the far-end temperature, and the collection frequency of the sensor is set to once every 30 seconds.

4. A mold production information data analysis method according to claim 1, characterized in that: The training process of the LSTM prediction model also includes: The first 70% of the historical normal temperature data of the mold heat treatment process is used as the training set; The middle 15% of historical normal temperature data is used as a validation set; The last 15% of the historical normal temperature data is used as the test set.

5. A mold production information data analysis system, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the mold production information data analysis method according to any one of claims 1 to 4 is implemented.

Images