Mold height adaptive adjustment method and system based on machine learning

Abstract

The present invention relates to the field of adaptive adjustment technology for mold height, and the method and system are as follows: collecting time series data within a preset time period and performing preprocessing to obtain a corresponding data sequence; dividing the data sequence to obtain a number of sequence segments, and performing linear system detection on each sequence segment; iteratively dividing the sequence segments of the nonlinear system based on the detection results to obtain a number of linear sequence segments; clustering all linear sequence segments to obtain multiple cluster clusters, and obtaining the optimal PID control parameters corresponding to each cluster cluster; collecting the operating data at the current moment in real time and matching the corresponding cluster clusters, and realizing adaptive dynamic adjustment of the mold height based on the corresponding optimal PID control parameters; the present invention converts the global nonlinear problem into local linear control through sequence division, which can solve the problem of insufficient precision of PID control caused by nonlinearity and realize adaptive adjustment of the mold height.

Description

Technical Field

[0001] The present invention relates to the field of adaptive adjustment technology, and in particular to a mold height adaptive adjustment method and system based on machine learning. Background Art

[0002] In traditional mold manufacturing and use, mold height adjustment typically relies on manual experience and operation. This method is not only inefficient but also difficult to precisely control, prone to errors caused by human factors, which in turn affects the stability and consistency of product quality. With the rapid development of industrial automation and intelligent manufacturing, the requirements for precision and efficiency in the production process are becoming increasingly stringent. As a key tool in the manufacturing industry, precise height adjustment of molds is crucial to improving production efficiency and ensuring product quality. Traditional adjustment methods are no longer able to meet the high-precision and high-efficiency requirements of modern industrial production.

[0003] During the use of high-speed presses, the height adjustment of the slider and the mold is a crucial link. The traditional adjustment method relies on a motor-driven gear set, which is achieved by connecting a worm gear assembly to a crank mechanism. However, this method has problems such as high power requirements, high costs, and is not conducive to lightweight design. To solve these problems, existing technologies have proposed installing cylinder balancing devices on both sides of the slider, using the cylinder to provide an upward balancing force to offset the slider's own weight, thereby reducing the driving force required by the motor. However, these technical solutions still lack intelligence and poor accuracy.

[0004] Machine learning technology has made significant progress in recent years. Its powerful capabilities in data analysis, prediction, and optimization offer new insights into solving complex industrial problems. Machine learning algorithms can model and analyze the various parameters and variables involved in mold height adjustment, enabling intelligent, adaptive adjustments. Summary of the Invention

[0005] In order to solve the problem of inaccurate control due to nonlinearity when adjusting mold height using the existing PID control method, and to achieve intelligent adaptive adjustment of the mold height of a high-speed press, the present invention provides a mold height adaptive adjustment method and system based on machine learning. The technical solution is as follows:

[0006] In a first aspect, the present invention provides a method for adaptively adjusting mold height based on machine learning, the method comprising the following steps: collecting time series data within a preset time period and performing preprocessing to obtain a corresponding data sequence; dividing the data sequence to obtain a plurality of sequence segments, and performing a linear system test on each sequence segment; iteratively dividing the sequence segments of the nonlinear system based on the linear system test results to obtain a plurality of linear sequence segments; clustering all the linear sequence segments to obtain a plurality of cluster clusters, constructing a PID control model based on each cluster cluster, and obtaining the optimal PID control parameters corresponding to each cluster cluster; collecting the current operating data in real time and matching it to the corresponding cluster cluster, and achieving adaptive dynamic adjustment of the mold height based on the corresponding optimal PID control parameters;

[0007] Among them, the data sequence includes a mold height difference sequence, a motor speed sequence, and a cylinder pressure sequence. The motor speed sequence and the cylinder pressure sequence are segmented respectively using an ordered sample clustering method or a threshold segmentation method to obtain a number of segmentation nodes. After superimposing the segmentation nodes of the two, all sequences are divided to obtain a number of sequence segments. Then, by respectively calculating the cross-correlation function between the sequence segments corresponding to the motor speed and cylinder pressure and the sequence segments corresponding to the mold height difference, or by detecting overlap and homogeneity, it is determined whether the corresponding sequence segment is a sequence segment of a nonlinear system.

[0008] Preferably, a sensor device for detecting the rotational speed is installed on the motor to collect the motor's rotational speed data in real time, a sensor device for detecting the pressure is installed in the cylinder or in the air path to collect the cylinder's pressure data in real time, a sensor device for detecting the height is installed on the mold of the high-speed press to collect the mold's height data in real time, and at the same time, the mold height set by the high-speed press is collected, the height difference between the set height and the actual height is calculated, and the mold height difference data is obtained in real time; the actual operation scenario of the high-speed press is simulated, and the motor speed, cylinder pressure, and mold height difference data are synchronously collected in real time within a preset time period to generate corresponding time series data; the timestamps of the time series data are aligned, the outliers in the data are deleted, and the missing values ​​are filled in before normalization processing is performed to obtain the mold height difference sequence, motor speed sequence, and cylinder pressure sequence.

[0009] Preferably, the ordered sample clustering method is used to cluster the element values ​​in the motor speed sequence and the cylinder pressure sequence respectively, and a number of cluster clusters and segmentation nodes are obtained respectively; or, according to the actual operation of the high-speed press, the threshold segmentation method is used to set the threshold parameters, and the motor speed and cylinder pressure are divided into multiple levels, and the motor speed sequence and the cylinder pressure sequence are segmented according to the divided levels to obtain a number of segmentation nodes; then the segmentation nodes of the motor speed sequence and the cylinder pressure sequence are superimposed, and the motor speed sequence, cylinder pressure sequence, and mold height difference sequence are divided by non-overlapping segmentation nodes to obtain a number of corresponding sequence segments.

[0010] Preferably, the cross-correlation function between each sequence segment corresponding to the motor speed and the mold height difference is calculated in turn. When there is a clear peak in the cross-correlation function corresponding to a certain sequence segment, the corresponding sequence segments conform to the linear system; similarly, it is determined whether the sequence segments of the cylinder pressure and the mold height difference conform to the linear system; or, the sequence segments corresponding to the motor speed and the cylinder pressure are respectively detected to determine whether the overlap and homogeneity are satisfied with the sequence segments of the corresponding mold height difference. If they are satisfied, it is determined that the corresponding sequence segments conform to the linear characteristics; when the sequence segments corresponding to both the motor speed and the cylinder pressure conform to the linear system with the sequence segments of the mold height difference at the same time, the corresponding sequence segment is determined to conform to the linear system. When any of the sequence segments corresponding to the motor speed and the cylinder pressure does not conform to the linear system, the corresponding sequence segment is determined to be a sequence segment of a nonlinear system.

[0011] Preferably, the division process of the motor speed sequence and the cylinder pressure sequence is repeated, and the sequence segments of the nonlinear system are divided more finely to obtain several new sequence segments. At the same time, the linear system detection process is repeated, and all new sequence segments are tested again until all the divided sequence segments meet the linear characteristics, thereby obtaining several linear sequence segments corresponding to the motor speed sequence, the cylinder pressure sequence, and the mold height difference sequence, respectively.

[0012] Preferably, a dynamic time warping algorithm is used to calculate the distances between similar linear sequence segments corresponding to the motor speed sequence, cylinder pressure sequence, and mold height difference sequence, and the sum of the distances is used as the similarity distance between the corresponding two linear sequence segments; based on the similarity distances between the linear sequence segments, all linear sequence segments are clustered to obtain several clusters corresponding to different working conditions.

[0013] Preferably, the linear sequence segments corresponding to the motor speed sequence, the cylinder pressure sequence, and the mold height difference sequence are converted from the time domain to the frequency domain using Fourier transform, and the Euclidean distances between each pair of linear sequence segments in the same sequence are calculated respectively; the sum of the Euclidean distances is used as the similarity distance between the corresponding two linear sequence segments; based on the similarity distances between the linear sequence segments, all the linear sequence segments are clustered to obtain a number of clusters corresponding to different working conditions.

[0014] Preferably, the operation process of the high-speed press is simulated according to the data element values ​​of the motor speed, cylinder pressure and mold height difference contained in the linear sequence segments of each cluster, and PID control models corresponding to different clusters are established respectively. The motor speed and cylinder pressure are matched according to the value of the mold height difference, and the various parameters of the PID control are gradually adjusted to adjust and control the mold height until the mold height difference is zero, thereby obtaining the optimal PID control parameters corresponding to each cluster.

[0015] Preferably, the motor speed data, cylinder pressure data, and mold height difference data at the current moment are collected in real time, and the distances between the data at the current moment and the centers of each cluster are calculated in turn. The data at the current moment are assigned to the cluster corresponding to the minimum distance value, so that the working conditions at the current moment match the working conditions corresponding to the cluster to which it belongs. The optimal PID control parameters corresponding to the cluster to which it belongs are called and used as the control parameters of the high-speed press at the next moment to adjust the mold height, thereby realizing adaptive dynamic adjustment of the mold height.

[0016] In second aspect, the present invention provides a mold height adaptive adjustment system based on machine learning, which is used to implement the above-mentioned mold height adaptive adjustment method, including: a processor, a memory, a communication interface and a sensor device deployed on a high-speed press, the communication interface is electrically connected to the sensor device and the electronic control system of the high-speed press, and the processor stores computer program instructions for implementing the mold height adaptive adjustment method.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] The present invention divides the motor speed sequence and the cylinder pressure sequence to convert the global nonlinear problem into local linear control, thereby solving the problem of inaccurate PID control caused by nonlinearity; then, by clustering the linear sequence segments, a number of clusters corresponding to different working conditions are obtained, and then the optimal PID control parameters corresponding to each cluster are obtained. At the same time, by clustering and merging similar working conditions, frequent parameter adjustments during the operation of the high-speed press can be avoided; after obtaining the optimal PID control parameters corresponding to different working conditions, the optimal PID control parameters corresponding to the current working condition can be matched in real time in the actual operation scenario of the high-speed press according to the data collected at the current moment, thereby realizing intelligent adaptive adjustment of the mold height of the high-speed press and improving the control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flowchart for implementing the mold height adaptive adjustment method;

[0020] Figure 2 This is the structural block diagram of the mold height adaptive adjustment system; DETAILED DESCRIPTION

[0021] The technical features of the present invention are further described in detail below with reference to the accompanying drawings so that those skilled in the art can understand them.

[0022] A method for adaptively adjusting mold height based on machine learning, the implementation process is as follows Figure 1 The specific implementation steps are as follows:

[0023] Step S1: collecting time series data within a preset period and preprocessing it to obtain a corresponding data sequence;

[0024] Specifically, a sensor device for detecting the rotational speed, such as a Hall sensor, an electromagnetic induction sensor, etc., is installed on the motor to collect the rotational speed data of the motor in real time; a sensor device for detecting pressure, such as a cylinder pressure sensor or a gas pressure sensor, is installed in the cylinder or in the air path to collect the pressure data of the cylinder in real time; a sensor device for detecting the height, such as a high-precision laser ranging sensor, is installed on the mold of the high-speed press to collect the height data of the mold in real time, and at the same time collect the mold height set by the high-speed press, calculate the height difference between the set height and the actual height, and obtain the mold height difference data in real time; simulate the actual operation scenario of the high-speed press, and synchronously collect the motor rotational speed, cylinder pressure, and mold height difference data in real time within a preset time period to generate corresponding time series data; align the timestamps of the time series data, delete the outliers in the data, fill in the missing values, and then perform normalization processing; the longer the preset time period, the wider the working conditions for simulating the control of the mold height, the more data samples, the more comprehensive the corresponding working conditions when the sequence is subsequently divided, and the more accurate the control effect;

[0025] Among them, the Z-Score method is used to detect and delete outliers in the time series data, and then the missing values ​​are filled by linear interpolation. Then, the ratios of the data values ​​of motor speed, cylinder pressure, and mold height difference to the maximum values ​​in the corresponding data sequences are used as the corresponding normalized values ​​to obtain the normalized mold height difference sequence, motor speed sequence, and cylinder pressure sequence.

[0026] Step S2: Divide the data sequence into several sequence segments, and perform linear system detection on each sequence segment;

[0027] Controlling mold height is a dynamic process, and traditional dynamic control methods mostly use PID control. When the relationship between the control variable and the system response is nonlinear, that is, the relationship between the PID output and the output of the controlled object is nonlinear, the input-output relationship does not satisfy the superposition principle. Problems such as saturation and dead zone may exist, which can easily lead to a decrease in PID control accuracy. Conversely, when the relationship between the two conforms to a linear system, the PID control adjustment effect is better and the control accuracy is higher. Therefore, in order to achieve better control effect and higher accuracy of mold height, it is necessary to divide the sequence data corresponding to the control variable and the system response into different sequence segments, ensuring that the relationship between the corresponding sequence segments of the two conforms to a linear system, thereby converting the global nonlinear problem into local linear control.

[0028] The first embodiment of the sequence division method is as follows:

[0029] The ordered sample clustering method is used to cluster the element values ​​in the motor speed sequence and cylinder pressure sequence respectively, and several cluster clusters and segment nodes are obtained respectively. After superimposing the segment nodes of the motor speed sequence and cylinder pressure sequence, the motor speed sequence, cylinder pressure sequence, and mold height difference sequence are divided by non-overlapping segment nodes to obtain several sequence segments; the ordered sample clustering method can retain the temporal order, which is convenient for the subsequent superposition of segment nodes.

[0030] The second embodiment of the sequence division method is as follows:

[0031] According to the actual operation of the high-speed press, the threshold parameters are set by the threshold segmentation method, and the motor speed and cylinder pressure are divided into multiple levels, such as high speed, medium speed and low speed, and the cylinder pressure is divided into high pressure, medium pressure and low pressure. The motor speed sequence and cylinder pressure sequence are segmented according to the divided levels, and then the segmentation nodes of the motor speed sequence and cylinder pressure sequence are superimposed. The motor speed sequence, cylinder pressure sequence and mold height difference sequence are divided by non-overlapping segmentation nodes to obtain several sequence segments.

[0032] The first embodiment of the linear system detection method is as follows:

[0033] Separately detect whether the sequence segments corresponding to the motor speed and cylinder pressure and the sequence segments corresponding to the mold height difference satisfy superposition and homogeneity. Superposition means that the system response caused by the joint action of multiple control quantities is equal to the sum of the system responses caused by each control quantity acting alone. Homogeneity means that when only one control quantity acts, the system response is proportional to the control quantity, wherein the control quantity is the sequence segment data corresponding to the motor speed and cylinder pressure, and the system response is the sequence segment data corresponding to the mold height difference. When the sequence segments corresponding to the motor speed and the mold height difference, and the sequence segments corresponding to the cylinder pressure and the mold height difference satisfy superposition and homogeneity at the same time, it is determined that the corresponding sequence segment meets the linear characteristics. In addition, the corresponding sequence segment is determined to be a sequence segment of a nonlinear system.

[0034] The second embodiment of the linear system detection method is as follows:

[0035] Since the cross-correlation function can measure the similarity of two time series signals and the peak position reflects the lag relationship, linear dependence can be detected by calculating the cross-correlation function; the cross-correlation function between each motor speed and mold height difference corresponding sequence segment is calculated in turn. When the cross-correlation function corresponding to a certain sequence segment has a clear peak, the corresponding sequence segment between the motor speed and mold height difference conforms to a linear system; similarly, the cross-correlation function between each cylinder pressure and mold height difference corresponding sequence segment is calculated in turn. When the cross-correlation function corresponding to a certain sequence segment has a clear peak, the corresponding sequence segment between the cylinder pressure and mold height difference conforms to a linear system; when the sequence segments corresponding to both the motor speed and cylinder pressure conform to the linear system with the sequence segment of the mold height difference at the same time, the corresponding sequence segment is judged to conform to the linear system. When any of the sequence segments corresponding to the motor speed and cylinder pressure does not conform to the linear system, the corresponding sequence segment is judged to be a sequence segment of a nonlinear system.

[0036] Step S3: Based on the linear system detection results, iteratively divide the sequence segments of the nonlinear system to obtain several linear sequence segments;

[0037] When a sequence segment conforms to a linear system, the sequence segment does not need to be further divided; conversely, when a sequence segment does not conform to a linear system, the sequence segment needs to be further divided, and the division method and process are consistent with step S2; similarly, after the division, each newly divided sequence segment needs to be tested again to see if it conforms to the linear system, and the linear system detection method and process are consistent with step S2.

[0038] Specifically, the ordered sample clustering method or the threshold segmentation method is used to repeat the division process of the motor speed sequence and the cylinder pressure sequence, and the sequence segments of the nonlinear system are divided more finely to obtain several new sequence segments. At the same time, the linear system detection process is repeated, and the linear system detection is performed again on all new sequence segments until all the divided sequence segments meet the linear characteristics, and several linear sequence segments corresponding to the motor speed sequence, cylinder pressure sequence, and mold height difference sequence are obtained.

[0039] Step S4: clustering all linear sequence segments to obtain multiple clusters, building a PID control model based on each cluster, and obtaining the optimal PID control parameters corresponding to each cluster;

[0040] After the sequence is divided, several linear sequence segments are obtained. The linear sequence segments corresponding to different time periods may have similar working conditions. Therefore, the linear sequence segments are clustered and similar working conditions are merged through clustering to avoid frequent parameter adjustments during the operation of the high-speed press.

[0041] The first embodiment of the distance calculation method in the clustering process is as follows:

[0042] The dynamic time warping algorithm is used to calculate the distances between the same linear sequence segments corresponding to the motor speed sequence, cylinder pressure sequence, and mold height difference sequence. Specifically, the distance calculation formula between the same linear sequence segments is:

[0043]

[0044] Where, represents the distance between the i-th linear sequence segment and the j-th linear sequence segment in the same sequence, represents the i-th linear sequence segment, represents the jth linear sequence segment, represents the dynamic time warping algorithm; then, the sum of the corresponding distances in the three types of sequences is taken as the similarity distance between the corresponding two linear sequence segments.

[0045] The second embodiment of the distance calculation method in the clustering process is as follows:

[0046] The linear sequence segments corresponding to the motor speed sequence, cylinder pressure sequence, and mold height difference sequence are converted from the time domain to the frequency domain using Fourier transform, and the Euclidean distance between each pair of linear sequence segments in the same sequence is calculated. Specifically, the Euclidean distance calculation formula between each pair of linear sequence segments in the same sequence is:

[0047]

[0048] Where, represents the Euclidean distance between the i-th linear sequence segment and the j-th linear sequence segment in the same sequence, represents the spectrum of the kth frequency in the i-th linear sequence segment, represents the spectrum of the kth frequency in the jth linear sequence segment, It represents the number of frequencies after the linear sequence segment is converted to the frequency domain. Afterwards, the sum of the corresponding Euclidean distances in the three types of sequences is also taken as the similarity distance between the corresponding two linear sequence segments.

[0049] After obtaining the similarity distances between each pair of linear sequence segments, all linear sequence segments are clustered using a k-means clustering method or a density-based clustering method based on the similarity distances between the linear sequence segments to obtain a number of clusters corresponding to different working conditions.

[0050] The optimal PID control parameters corresponding to each cluster are obtained. The specific process is as follows:

[0051] According to the data element values ​​of motor speed, cylinder pressure and mold height difference contained in the linear sequence segments of each cluster, the operation process of the high-speed press is simulated, and a PID control model is constructed for each cluster. The PID control parameters include proportional adjustment parameters, integral adjustment parameters and differential adjustment parameters. The motor speed and cylinder pressure are matched according to the value of the mold height difference, and the various PID control parameters are gradually adjusted to adjust the mold height until the mold height difference is zero. Based on the PID control model and the existing PID control parameter calculation method, the optimal PID control parameters corresponding to each PID control model are obtained. The PID control model is associated with the cluster one by one, and the corresponding optimal PID control parameters are matched for each cluster to achieve real-time matching of the optimal PID control parameters corresponding to different working conditions.

[0052] Step S5: real-time collection of current operating data and matching of the corresponding clusters, and adaptive dynamic adjustment of the mold height based on the corresponding optimal PID control parameters;

[0053] Specifically, the motor speed data, cylinder pressure data, and mold height difference data at the current moment are collected in real time, and the distance between the data at the current moment and the center of each cluster is calculated in turn. The data at the current moment is assigned to the cluster corresponding to the minimum distance value, so that the working conditions at the current moment match the working conditions corresponding to the cluster to which it belongs. The optimal PID control parameters corresponding to the cluster to which it belongs are called as the control parameters of the high-speed press at the next moment to adjust the mold height, thereby realizing adaptive dynamic adjustment of the mold height.

[0054] The present invention also discloses a mold height adaptive adjustment system based on machine learning, the structure of which is as follows: Figure 2 As shown, the method for implementing the above-mentioned mold height adaptive adjustment method based on machine learning includes: a processor, a memory, a communication interface and a sensor device deployed on a high-speed press. The communication interface is electrically connected to the sensor device and the electronic control system of the high-speed press. The processor stores computer program instructions for implementing the above-mentioned mold height adaptive adjustment method, the memory is used to store data, and the communication interface is used to receive and transmit data.

[0055] The embodiments included in the present invention are merely descriptions of preferred implementation methods of the present invention and are not limited to the precise structures described above and shown in the accompanying drawings. Various modifications and changes can be made without departing from the scope of protection of the present invention. Without departing from the design concept of the present invention, various variations and improvements made to the technical solutions of the present invention by engineers and technicians in this field should fall within the scope of protection of the present invention.

Claims

1. A method for adaptively adjusting mold height based on machine learning, characterized by: Time series data within a preset time period is collected and preprocessed to obtain a corresponding data sequence; the data sequence is divided into several sequence segments, and a linear system test is performed on each sequence segment; based on the linear system test results, the sequence segments of the nonlinear system are iteratively divided to obtain several linear sequence segments; Cluster all linear sequence segments to obtain multiple clusters, build a PID control model based on each cluster, and obtain the optimal PID control parameters corresponding to each cluster; Real-time collection of current operating data and matching of the corresponding clusters to achieve adaptive dynamic adjustment of mold height based on the corresponding optimal PID control parameters; The method collects mold height data in real time, and simultaneously collects the mold height set by the high-speed press, calculates the height difference between the set height and the actual height, and obtains mold height difference data in real time. The data sequence includes a mold height difference sequence, a motor speed sequence, and a cylinder pressure sequence. The motor speed sequence and the cylinder pressure sequence are segmented using an ordered sample clustering method or a threshold segmentation method to obtain a number of segment nodes. After superimposing the segment nodes of the two, all sequences are divided to obtain a number of sequence segments. Subsequently, the cross-correlation function between the sequence segments corresponding to the motor speed and cylinder pressure and the sequence segment corresponding to the mold height difference is calculated, or by detecting overlap and homogeneity, it is determined whether the corresponding sequence segment is a sequence segment of a nonlinear system. Repeat the division process of the motor speed sequence and the cylinder pressure sequence, perform a finer division operation on the sequence segments of the nonlinear system, and obtain several new sequence segments. At the same time, repeat the linear system detection process, and perform linear system detection on all new sequence segments again until all divided sequence segments meet the linear characteristics, and obtain several linear sequence segments corresponding to the motor speed sequence, cylinder pressure sequence, and mold height difference sequence respectively.

2. The method for adaptively adjusting mold height based on machine learning according to claim 1, characterized in that: The method collects time series data within a preset time period and performs preprocessing, including: installing a sensor device for detecting the rotational speed on the motor to collect the motor rotational speed data in real time, installing a sensor device for detecting the pressure in the cylinder or in the air path to collect the cylinder pressure data in real time, installing a sensor device for detecting the height on the mold of the high-speed press to obtain the mold height difference data in real time; simulating the actual operation scenario of the high-speed press, synchronously collecting the motor rotational speed, cylinder pressure, and mold height difference data in real time within a preset time period to generate corresponding time series data; aligning the timestamps of the time series data, deleting outliers in the data, filling in missing values, and then performing normalization processing to obtain a mold height difference sequence, a motor rotational speed sequence, and a cylinder pressure sequence.

3. The method for adaptively adjusting mold height based on machine learning according to claim 1, characterized in that: The data sequence is divided into several sequence segments, including: using an ordered sample clustering method to cluster the element values ​​in the motor speed sequence and the cylinder pressure sequence respectively, to obtain several cluster clusters and segmentation nodes respectively; or, according to the actual operation of the high-speed press, using a threshold segmentation method to set threshold parameters, dividing the motor speed and cylinder pressure into multiple levels, and segmenting the motor speed sequence and the cylinder pressure sequence according to the divided levels to obtain several segmentation nodes; then superimposing the segmentation nodes of the motor speed sequence and the cylinder pressure sequence, and dividing the motor speed sequence, the cylinder pressure sequence, and the mold height difference sequence through non-overlapping segmentation nodes to obtain several mutually corresponding sequence segments.

4. The method for adaptively adjusting mold height based on machine learning according to claim 1, characterized in that: The linear system detection is performed on each sequence segment, including: calculating the cross-correlation function between each sequence segment corresponding to the motor speed and the mold height difference in sequence, when the cross-correlation function corresponding to a certain sequence segment has a clear peak, the corresponding sequence segments conform to the linear system; similarly, determining whether the sequence segments of the cylinder pressure and the mold height difference conform to the linear system; or, respectively detecting whether the sequence segments corresponding to the motor speed and the cylinder pressure satisfy the superposition and homogeneity with the sequence segments corresponding to the mold height difference, if so, determining that the corresponding sequence segments conform to the linear characteristics; when the sequence segments corresponding to both the motor speed and the cylinder pressure and the sequence segments corresponding to the mold height difference conform to the linear system at the same time, determining that the corresponding sequence segment conforms to the linear system; when either of the sequence segments corresponding to the motor speed and the cylinder pressure does not conform to the linear system, determining that the corresponding sequence segment is a sequence segment of a nonlinear system.

5. The method for adaptively adjusting mold height based on machine learning according to claim 1, characterized in that: The method of clustering all linear sequence segments to obtain multiple clusters includes: using a dynamic time warping algorithm to calculate the distances between similar linear sequence segments corresponding to the motor speed sequence, the cylinder pressure sequence, and the mold height difference sequence, and taking the sum of the distances as the similarity distance between the corresponding two linear sequence segments; and clustering all linear sequence segments based on the similarity distances between the linear sequence segments to obtain a number of clusters corresponding to different working conditions.

6. The method for adaptively adjusting mold height based on machine learning according to claim 1, characterized in that: The method of clustering all linear sequence segments to obtain multiple clusters further includes: using Fourier transform to convert the linear sequence segments corresponding to the motor speed sequence, the cylinder pressure sequence, and the mold height difference sequence from the time domain to the frequency domain, and respectively calculating the Euclidean distances between each pair of linear sequence segments in the same sequence; using the sum of the Euclidean distances as the similarity distance between the corresponding two linear sequence segments; and clustering all linear sequence segments based on the similarity distances between the linear sequence segments to obtain a plurality of clusters corresponding to different working conditions.

7. The method for adaptively adjusting mold height based on machine learning according to claim 5 or 6, characterized in that: The method constructs a PID control model based on each cluster to obtain the optimal PID control parameters corresponding to each cluster, including: simulating the operation process of the high-speed press according to the data element values ​​of the motor speed, cylinder pressure, and mold height difference contained in the linear sequence segments of each cluster, establishing PID control models corresponding to different clusters, matching the motor speed and cylinder pressure according to the value of the mold height difference, and gradually adjusting various PID control parameters to adjust and control the mold height until the mold height difference is zero, thereby obtaining the optimal PID control parameters corresponding to each cluster.

8. The method for adaptively adjusting mold height based on machine learning according to claim 7, characterized in that: The adaptive dynamic adjustment of mold height based on corresponding optimal PID control parameters includes: real-time collection of motor speed data, cylinder pressure data, and mold height difference data at the current moment, sequentially calculating the distance between the data at the current moment and the center of each cluster, assigning the data at the current moment to the cluster corresponding to the minimum distance value, so that the working conditions at the current moment match the working conditions corresponding to the cluster to which it belongs, calling the optimal PID control parameters corresponding to the cluster to which it belongs as the control parameters of the high-speed press at the next moment to adjust the mold height, thereby realizing adaptive dynamic adjustment of the mold height.

9. A mold height adaptive adjustment system based on machine learning, characterized by: The invention comprises a processor, a memory, a communication interface and a sensor device deployed on a high-speed press, wherein the communication interface is electrically connected to the sensor device and the electronic control system of the high-speed press, and the processor stores computer program instructions for implementing the mold height adaptive adjustment method according to any one of claims 1 to 8.

Images