A method and system for controlling and early warning of a forging production

Abstract

The present application belongs to the technical field of management and control early warning, and particularly relates to a kind of management and control early warning method and system for forging production, and the method thereof comprises: collecting the data value of forging press in multiple monitoring dimensions;Calculate the local distortion performance of target time in each dimension within the time window;Fusion local distortion and global distortion degree, calculate the sensitivity of target time in each dimension;According to the sensitivity, adjust the initial cutoff distance, and construct an adaptive neighborhood sphere for the target time;Finally, based on the adaptive neighborhood sphere of each time within the time window, density peak clustering is carried out, and according to the deviation of the target time sample point, the sub-health state is identified and early warning is generated.The present application solves the problem of traditional density peak clustering algorithm that high fluctuation signal covers low fluctuation key signal by constructing an adaptive neighborhood sphere, significantly improving the identification accuracy of early failure of equipment in the forging production process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of management and control early warning. More particularly, the present application relates to a management and control early warning method and system for forging production. BACKGROUND

[0002] In high-end manufacturing industries such as automobiles and aerospace, forgings as core load-bearing components have extremely strict quality requirements, and the running stability of the production equipment, i.e., the forging press, is the lifeline to ensure consistent quality. Currently, the monitoring of the forging press in the industry generally relies on the alarm system based on fixed process parameter thresholds provided by the equipment.

[0003] However, this monitoring mode has significant defects. The forging press does not suddenly change from healthy to failure, but rather goes through a sub-healthy stage of slow performance degradation. Although the small changes in equipment parameters during this stage do not trigger the fixed threshold alarm, they are enough to cause quality problems such as substandard forging size precision or internal micro-cracks. Therefore, the existing early warning method cannot identify this sub-healthy state.

[0004] Considering that the distribution density of process parameters changes when the equipment evolves from a healthy state to a sub-healthy state, the data points in the sub-healthy state usually exhibit lower local density compared to the data points in the healthy state. Therefore, the density peak clustering algorithm (DPC) can be used to distinguish between normal data clustering structures and abnormal data points. However, the traditional DPC algorithm uses a fixed cutoff distance, which is equivalent to constructing a fixed hypersphere around each data point as its neighborhood in a multi-dimensional space. This one-size-fits-all approach to all dimension features leads to the analysis results being excessively influenced by high-variation dimension features with a larger variation range, resulting in inaccurate identification of the sub-healthy state of the equipment and missed detection. SUMMARY

[0005] To solve the technical problem of inaccurate identification of the sub-healthy state of the equipment and easy masking by high-variation features due to the use of a fixed cutoff distance for all dimension features in the traditional density peak clustering algorithm, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a management and control early warning method for forging production, comprising: collecting data values in multiple monitoring dimensions at several time points during the production process of the forging press; determining the local distortion performance of each monitoring dimension at the current time point according to the difference between the data distribution of each monitoring dimension within the time window of the current time point; and determining the sensitivity of each monitoring dimension at the current time point according to the local distortion performance of each monitoring dimension at the current time point and the global distortion performance in all monitoring dimensions.

[0007] The initial cutoff distance at the current moment is adjusted based on sensitivity to obtain the adjusted cutoff distance at the current moment in each monitoring dimension. Based on the cutoff distance at the current moment in each monitoring dimension, a neighborhood sphere is constructed for the current moment. Sample points from all moments within the current moment's time window are used as clustering samples, and the neighborhood sphere at each moment is used as the neighborhood range for clustering to obtain several clusters. Based on the deviation of the sample point at the current moment from its respective cluster, the operating status of the equipment at the current moment is determined, and control and early warning are issued.

[0008] This invention overcomes the fundamental limitation of fixed and symmetrical neighborhoods in traditional density peak clustering algorithms by dynamically calculating the sensitivity of each monitoring dimension and constructing a neighborhood sphere of the equipment's real-time status based on this. This method enables the shape and range of the neighborhood to automatically shrink in potential abnormal dimensions, thereby solving the technical problem that subtle anomalies in low-fluctuation dimensions are masked by high-fluctuation dimensions due to huge differences in the numerical range between data dimensions. In addition, the sensitivity calculation integrates the local deviation of a single dimension and the global risk of the equipment as a whole, making the status assessment more comprehensive and robust. From data acquisition, feature construction, neighborhood adjustment to final anomaly identification, a closed-loop control and early warning process is formed, significantly reducing the missed detection rate and thus ensuring the production quality of forging products.

[0009] Preferably, the local distortion behavior in each monitoring dimension at the current moment satisfies the expression: In the formula, Indicates the current time at the th order. Local distortion characteristics of each monitoring dimension; Indicates the current time at the th order. Data values ​​for each monitoring dimension; This indicates that within the current time window, all other times are in the [number]th [time window]. The mean and standard deviation of the data values ​​for each monitoring dimension; Indicates the preset hyperparameters; Indicates the absolute value sign; This represents the standard normalization function.

[0010] This invention effectively obtains the distortion of each monitoring dimension at the current moment by calculating the standardized deviation between the data value of each monitoring dimension at the current moment and the mean of the remaining data within the time window. This method is not affected by the changes in the numerical range of each dimension's data, and provides data support for subsequent sensitivity calculations.

[0011] Preferably, the sensitivity of the current moment in each monitoring dimension satisfies the expression: In the formula, Indicates the current time at the th order. Sensitivity of each monitoring dimension; Indicates the current time at the th order. Local distortion characteristics of each monitoring dimension; This indicates the global distortion performance across all monitoring dimensions at the current moment; This represents the standard normalization function.

[0012] This invention integrates the local distortion performance of a single dimension with the global distortion performance of all dimensions to dynamically calculate the sensitivity of each dimension. By introducing a global risk adjustment coefficient, the system can maintain a higher level of vigilance to small changes in all dimensions when the overall state of the equipment is unstable, thereby achieving a comprehensive assessment of risk and improving the robustness of the early warning model and its ability to capture systemic risks.

[0013] Preferably, the global distortion performance across all monitoring dimensions at the current moment is equal to the average of the local distortion performance across all monitoring dimensions at the current moment.

[0014] Preferably, the truncation distance adjusted for each monitoring dimension at the current moment satisfies the expression: In the formula, Indicates the current time at the th order. Cutoff distance after adjustment of each monitoring dimension; This represents the initial cutoff distance at the current moment; Indicates the current time at the th order. Sensitivity of each monitoring dimension; This represents the natural exponential function.

[0015] This invention utilizes the natural exponential function to transform the calculated dynamic sensitivity into a nonlinear decay adjustment of the initial cutoff distance. This method ensures that the higher the sensitivity of a dimension, the more significantly its neighborhood search range shrinks, thereby achieving attention to potential anomalous dimensions in geometric space. This provides a crucial basis for subsequent clustering algorithms to accurately identify subtle anomalies masked by high-fluctuation dimensions.

[0016] Preferably, the initial cutoff distance at the current moment is equal to the Euclidean distance between the sample point at the current moment and the sample point at the current health reference moment; the current health reference moment is obtained by traversing the global distortion performance of each moment within the time window of the current moment across all monitoring dimensions and obtaining the moment when the global distortion performance is at its minimum.

[0017] Preferably, the step of constructing a neighborhood sphere at the current moment based on the truncation distance adjusted for each monitoring dimension at the current moment includes: constructing a multi-dimensional coordinate system with each monitoring dimension as the coordinate axis; taking the sample point at the current moment as the geometric center of the neighborhood sphere; starting from the geometric center, taking the truncation distance adjusted for the corresponding monitoring dimension along each coordinate axis as the radius of the neighborhood sphere along each coordinate axis, thus constructing the neighborhood sphere.

[0018] This invention constructs an asymmetric neighborhood sphere in a multidimensional space by using each monitoring dimension as a reference axis and its respective adaptive cutoff distance as a radius. This method breaks through the limitation of the globally fixed symmetric neighborhood sphere in traditional clustering algorithms, enabling the shape of the neighborhood to match the stable state differences of the device in different dimensions in real time. This allows for more accurate delineation of truly similar healthy samples, providing a reliable spatial division basis for accurately identifying abnormal sample points in a sub-healthy state.

[0019] Preferably, the clustering algorithm is the density peak clustering algorithm.

[0020] Preferably, the step of determining the current device operating status and issuing a control warning based on the deviation of the current sample point from its respective cluster includes: the deviation is measured by the Euclidean distance between the current sample point and the cluster center of its cluster. Is it greater than the average Euclidean distance between the other sample points in the same cluster and the cluster center? Certain: When Greater than When the device is in a sub-healthy state, a control warning is issued.

[0021] Secondly, the present invention provides a control and early warning system for forging production, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned control and early warning method for forging production is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned control and early warning method for forging production, and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows:

[0024] (1) By calculating the local deviation of the data values ​​of each monitoring dimension relative to its recent dynamic baseline, the present invention accurately captures the instantaneous and weak changes that are easily ignored by traditional trend analysis, thereby improving the sensitivity of early fault monitoring dimensions hidden under stable operating conditions.

[0025] (2) The application fuses local and global distortion performance for dynamic evaluation, can automatically amplify the sensitivity to changes in each monitoring dimension, and effectively identifies complex sub-health states with unobvious characteristics caused by the coordinated degradation of multiple components;

[0026] (3) The application solves the problem of missing detection of low fluctuation key signals hidden by high fluctuation signals by constructing an adaptive neighborhood sphere for each time, and improves the accuracy of the algorithm for abnormal state early warning. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a flowchart schematically showing a control and early warning method for a forging production in the application. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of, rather than all of, the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0029] The specific implementation manners of the application will be described in detail below with reference to the drawings.

[0030] The embodiments of the application disclose a control and early warning method for a forging production, with reference to Figure 1 , comprising steps S1-S5:

[0031] S1, collecting data values in multiple monitoring dimensions at several time points in a forging press production process.

[0032] It should be noted that in the forging press production, the multi-dimensional process data such as pressure, displacement and vibration corresponding to the equipment health state follow a stable and highly repeatable dynamic mode in each production cycle. Once early performance degradation of the equipment occurs, for example, the hydraulic system response is delayed or the mechanical components are abnormally worn, the dynamic mode of these process parameters will deviate slightly. If this deviation is not identified in time, it will lead to quality defects such as unqualified size precision of forgings, internal micro-cracks or decreased mechanical properties, and even cause accelerated damage to the equipment itself. Therefore, in order to accurately identify early abnormalities caused by equipment performance degradation, complete process data of each state parameter of the equipment in the forging production needs to be collected at a high frequency and synchronously, to provide a data basis for subsequent construction of an identification benchmark for sub-health states.

[0033] Specifically, a pressure sensor is installed on a main cylinder of a hydraulic system of the forging press to capture real-time pressure data applied to the workpiece during the downward movement of the slide; a displacement sensor is installed at the slide guide to monitor real-time displacement data of the slide; and a vibration sensor is installed at the hydraulic main rod area to capture real-time vibration data generated due to impact during the forging process; for example, the collection frequency of each sensor can be set to 20 Hz, and the implementer can adjust it according to the actual situation.

[0034] The data value of each monitoring dimension at each time point needs to be normalized to obtain the data value of each monitoring dimension at each time point; this step can eliminate the interference of unit dimension and order of magnitude difference on subsequent data clustering, and ensure that the data value of each monitoring dimension is analyzed on the same order of magnitude; for example, the normalization processing includes but is not limited to maximum and minimum normalization, linear normalization, etc.

[0035] At this point, the data value of each monitoring dimension at several time points is obtained.

[0036] S2, according to the difference between the data distribution of each monitoring dimension in the time window of the current time, determine the local distortion performance of each monitoring dimension at the current time.

[0037] It should be noted that the forging equipment enters the sub-health stage, which may be that the data value of a certain independent dimension deviates significantly, or the data values of multiple dimensions deviate at the same time. However, the change of the data value of each dimension is often different due to the difference between the data values of the dimensions, for example, the numerical range of the pressure signal can be as high as tens of megapascals, and the deviation value is also large; while the vibration signal collected at the same time, the numerical range may be only a few gravity accelerations, and the significant deviation value is also small; therefore, each dimension at each time point needs to be analyzed separately, and the distortion performance of each dimension at each time point is quantified based on the local change of the data value of each dimension itself, to prepare for the subsequent calculation of dynamic sensitivity.

[0038] Specifically, a time window is established for each time point, and each time point is the last time point of the time window in which the time point is located. For example, the size of the time window can be set to 30 seconds, and the implementer can adjust it according to the actual situation.

[0039] According to the difference between the data distribution of each monitoring dimension in the time window of the current time, determine the local distortion performance of each monitoring dimension at the current time; the local distortion performance satisfies the expression:

[0040]

[0041] In the formula, represents the local distortion performance of the i-th monitoring dimension at the current time; ​represents the data value of the current time in the first monitoring dimension; represents the average value and standard deviation of the data value of the current time in the first monitoring dimension within the time window of the current time and all other times; represents a preset hyperparameter for preventing the denominator from being 0 to cause the fraction to be meaningless, which can be set to 0.01; represents the absolute value symbol; represents the standard normalization function.

[0042] wherein, reflects the difference between the data value of the current time in the first monitoring dimension and the average level of the data value of the current time in the first monitoring dimension within the time window of the current time and all other times, the larger the value, the more significantly the data value of the current time in the first monitoring dimension deviates from the average level of the data value of the current time in the first monitoring dimension within the time window of the current time and all other times, and the magnitude of this deviation far exceeds the normal fluctuation range of the data value of the current time in the first monitoring dimension within the time window of the current time and all other times, which means that the current time in the first monitoring dimension deviates from its normal state, that is, the greater the local distortion performance of the current time in the first monitoring dimension.

[0043] At this point, the local distortion performance of the current time in each monitoring dimension is obtained.

[0044] S3, according to the local distortion performance of the current time in each monitoring dimension and the global distortion performance in all monitoring dimensions, determine the sensitivity of the current time in each monitoring dimension.

[0045] It should be noted that once the device has an unhealthy problem, it will gradually affect other components, so that the local distortion performance of each dimension increases. If the global distortion degree at a certain time is high, it means that the device itself is on the edge of sub-health, at this time, any slight change in all dimensions of the device should be highly vigilant. Therefore, an index that can integrate global distortion and local deviation is needed to dynamically calculate a sensitivity for each dimension at each time, thereby improving the accuracy of device sub-health state recognition.

[0046] Specifically, according to the local distortion performance of the current time in each monitoring dimension and the global distortion performance in all monitoring dimensions, the sensitivity of the current time in each monitoring dimension is determined; wherein the global distortion performance of the current time in all monitoring dimensions is equal to the average value of the local distortion performance of the current time in all monitoring dimensions.

[0047] The sensitivity of each monitoring dimension at the current moment satisfies the expression:

[0048]

[0049] In the formula, Indicates the current time at the th order. Sensitivity of each monitoring dimension; Indicates the current time at the th order. Local distortion characteristics of each monitoring dimension; This indicates the global distortion performance across all monitoring dimensions at the current moment; This represents the standard normalization function.

[0050] In particular, considering the situation where a certain monitoring dimension deviates significantly from its normal state at any given moment, indicating a potential anomaly in the corresponding device component, the system should be more sensitive to changes in this monitoring dimension's data; therefore, through... Reflecting the current moment in the The basic sensitivity of each monitoring dimension; in addition, if the corresponding device is currently on the verge of a sub-healthy state, each monitoring dimension may have abnormalities in its corresponding device component. At this time, we should be highly vigilant about any slight changes in all monitoring dimensions; therefore, a global risk adjustment coefficient is introduced. ,like The larger the value, the more significantly the device deviates from its normal state across multiple monitoring dimensions at the current moment. This indicates that the overall state of the device is unstable at the current moment, and the corresponding global risk adjustment coefficient will be much greater than 1, thus increasing the sensitivity of the device across each monitoring dimension at the current moment. Conversely, when the overall state of the device is stable at the current moment, the corresponding global risk adjustment coefficient will approach 1, which is mainly determined by the local distortion of each monitoring dimension itself.

[0051] At this point, the sensitivity of each monitoring dimension at the current moment has been obtained.

[0052] S4. Adjust the initial cutoff distance at the current moment according to the sensitivity to obtain the adjusted cutoff distance at the current moment in each monitoring dimension; and construct the neighborhood sphere at the current moment based on the cutoff distance at the current moment in each monitoring dimension.

[0053] It should be noted that the density of process parameter distribution differs significantly between the healthy and sub-healthy states of actual equipment. Density peak clustering algorithms can be used to distinguish abnormal process parameters. However, this algorithm uses a fixed cutoff distance. In this scenario, multi-dimensional process parameters are represented by a spherical neighborhood for each sample point, the shape of which remains unchanged across all dimensions. This can lead to early anomalies in independent dimensions being masked by normal dimension data and missed, resulting in inaccurate early anomaly identification. Therefore, it is necessary to dynamically adjust the neighborhood based on the sensitivity of each monitoring dimension obtained above, so that it is not limited to a fixed spherical neighborhood, thereby improving the accuracy of early anomaly identification.

[0054] Specifically, it iterates through the global distortion performance of each moment in all monitoring dimensions within the current time window, and determines the moment with the minimum global distortion performance as the current health reference moment; it constructs sample points for each moment using the data values ​​of each moment in each monitoring dimension within the current time window; it calculates the initial cutoff distance for the current moment, which is equal to the Euclidean distance between the sample points of the current moment and the sample points of the current health reference moment; and it determines the adjusted cutoff distance for each monitoring dimension based on the sensitivity of the current moment in each monitoring dimension and the initial cutoff distance of the current moment.

[0055] The current truncation distance, adjusted for each monitoring dimension, satisfies the expression:

[0056]

[0057] In the formula, Indicates the current time at the th order. Cutoff distance after adjustment of each monitoring dimension; This represents the initial cutoff distance at the current moment; Indicates the current time at the th order. Sensitivity of each monitoring dimension; This represents the natural exponential function.

[0058] in, Reflects the initial cutoff distance at the current moment. The smaller the value, the healthier each monitoring dimension is at the current moment, indicating a lower systemic risk at that moment; in this case, if the value is in the [missing value]... The greater the sensitivity of the monitoring dimension, the higher the sensitivity of the sample points within the initial cutoff distance range. The more likely the data values ​​in each monitoring dimension are to be anomalies, the more necessary it is to reduce the risk of anomalies in the third monitoring dimension. The initial cutoff distance for the first monitoring dimension is used to ensure that subsequent clustering identifies anomalies in independent dimensions; conversely, if the initial cutoff distance for the first monitoring dimension is not specified, then... The lower the sensitivity of the monitoring dimension, the lower the sensitivity of the sample points within the initial cutoff distance. The more normal the data values ​​for each monitoring dimension are, the more important it is to maintain the original initial cutoff distance; that is... The closer it gets to 1, The closer it gets to 0.37, the more... much smaller ; The closer it gets to 0, The closer it gets to 1, the more... Approaching or equal to As for The larger the value, the higher the perceived systemic risk at the current moment. The larger the value itself, the more significant its attenuation adjustment on sensitive dimensions will be.

[0059] At this point, the cutoff distance after adjustment for each monitoring dimension at the current moment has been obtained.

[0060] It should be further explained that, since the sensitivity of each monitoring dimension is different, the corresponding cutoff distances are usually not equal. In other words, the boundaries of all cutoff distances may not form a standard hypersphere. Therefore, it is necessary to quantize the multi-monitoring dimension data in a spatial geometric system to determine the neighborhood sphere.

[0061] Specifically, a multi-dimensional coordinate system is constructed with each monitoring dimension as the reference axis. In the coordinate system, the sample point at the current moment is determined as the geometric center of the neighborhood sphere at the current moment. Starting from the geometric center, along each coordinate axis, the truncation distance after the corresponding dimension adjustment is taken as the radius of the neighborhood sphere at the current moment on each coordinate axis. The radii of each coordinate axis together constitute the neighborhood sphere at the current moment.

[0062] For example, if the geometric center of the neighborhood sphere at the current moment is (0.2, 0.9, 0.1), and the corresponding cutoff distance after adjusting the monitoring dimension is (0.81, 0.24, 0.81), then the neighborhood sphere function at the current moment is: The function defines a non-standard hypersphere, the mathematical construction of which is a well-known technique and will not be elaborated here.

[0063] S5. Take all sample points within the current time window as cluster samples, and cluster the neighborhood sphere of each time as the neighborhood range to obtain several clusters. Based on the deviation of the sample point at the current time from its own cluster, determine the current equipment operating status and issue control warnings.

[0064] It should be noted that by adaptive neighborhood sphere density peak clustering, the data change sensitive characteristics of each dimension can be effectively adapted, and the sample points with similar running states in the local window are dynamically divided into the same cluster. By comparing the distance between the sample point at the current moment and the cluster center in which it is located, and the average distance of other sample points in the cluster to the center, the deviation of the sample point at the current moment can be quantified, so that the potential sub-health state of the sample point can be accurately identified.

[0065] Specifically, all sample points at all moments in the time window at the current moment are taken as clustering samples, and the neighborhood sphere at each moment is taken as the neighborhood range, and the density peak clustering algorithm is used for clustering to obtain a plurality of clustering clusters. It should be noted that the present application is based on real-time processing of time sequence characteristics, that is, the neighborhood sphere at each moment in the time window at the current moment has been obtained. The density peak clustering process is a known technology, and will not be described here.

[0066] For the clustering cluster in which the sample point at the current moment is located, the Euclidean distance between the sample point at the current moment and the cluster center of the clustering cluster is obtained , and the average Euclidean distance between the remaining sample points in the clustering cluster and the cluster center ; if is greater than , it is considered that the device running state at the current moment has deviated from the stable working condition category in which it is located, showing significant abnormal characteristics, and it is considered that the device has entered a sub-health state.

[0067] A control unit is provided, which controls the alarm system according to the real-time monitoring result, and the alarm system sends an alarm signal to notify the relevant personnel of the production line to implement emergency measures. When the device enters a sub-health state at the current moment, the control unit immediately controls the alarm system to send a flashing light.

[0068] Thus, the control and early warning of the forging production is completed.

[0069] The embodiment of the present application also discloses a control and early warning system for forging production, comprising a processor and a memory, and the memory stores computer program instructions, which realize the control and early warning method for forging production according to the present application when executed by the processor.

[0070] The above system also includes a communication bus and a communication interface and other components familiar to those skilled in the art, the setting and function of which are known in the art, and thus will not be described here.

Claims

1. A method for controlling and early warning of forgings production, characterized in that, The method comprises: collecting data values of a plurality of monitoring dimensions at several time points in a production process of a forging press; determining local distortion performance of each monitoring dimension at the current time point according to a difference between a data distribution of each monitoring dimension within a time window of the current time point, and satisfying an expression: ； In the formula, represents the local distortion performance of the current time in the first monitoring dimension; represents the data value of the current time in the first monitoring dimension; represents the average value and standard deviation of the data value of all other times in the first monitoring dimension within the time window of the current time; represents a preset hyperparameter; represents the absolute value symbol; represents the standard normalization function; determining a sensitivity of each monitoring dimension at the current time point according to the local distortion performance of each monitoring dimension at the current time point and the global distortion performance of all monitoring dimensions, and satisfying an expression: ； wherein, represents the sensitivity of the current time instant in the j-th monitoring dimension; represents the local distortion performance of the current time instant in the j-th monitoring dimension; represents the global distortion performance of the current time instant in all monitoring dimensions; represents the local distortion performance of the current time instant in all monitoring dimensions; represents the global distortion performance of the current time instant in all monitoring dimensions; represents a standard normalization function; the global distortion performance of the current time instant in all monitoring dimensions is equal to the average of the local distortion performance of the current time instant in all monitoring dimensions; adjusting an initial cut-off distance of the current time point according to the sensitivity to obtain an adjusted cut-off distance of each monitoring dimension at the current time point, and satisfying an expression: ； In the formula, represents the current time adjusted truncated distance in the first monitoring dimension; represents the initial truncated distance at the current time; represents the sensitivity of the current time in the first monitoring dimension; represents the natural exponential function; constructing a neighborhood sphere of the current time point based on the cut-off distance of each monitoring dimension at the current time point. The sample points of all time moments in a time window of a current time moment are taken as clustering samples, and the neighborhood sphere of each time moment is taken as a neighborhood range to cluster, to obtain a plurality of clustering clusters, the device running state of the current time moment is judged according to the deviation of the sample point of the current time moment in the clustering cluster to which it belongs, and control warning is performed.

2. The method according to claim 1, wherein, The initial cut-off distance of the current time point is equal to the Euclidean distance between the sample point of the current time point and the sample point of a healthy reference time point of the current time point. The healthy reference time point of the current time point is obtained by traversing each time point within a time window of the current time point, obtaining the global distortion performance of all monitoring dimensions, and obtaining the time point with the minimum global distortion performance.

3. The method according to claim 1, wherein The neighborhood sphere of the current time point is constructed based on the adjusted cut-off distance of each monitoring dimension at the current time point, comprising: constructing a multi-dimensional coordinate system with each monitoring dimension as a coordinate axis; taking the sample point of the current time point as the geometric center of the neighborhood sphere, and taking the adjusted cut-off distance of each coordinate axis as the radius of the neighborhood sphere along each coordinate axis to construct the neighborhood sphere.

4. The method of claim 1, wherein the method comprises: The clustering algorithm is a density peak clustering algorithm.

5. The method of claim 1, wherein the method comprises: The device running state of the current time point is determined according to the deviation of the sample point of the current time point in the cluster to which it belongs, and control and early warning are performed, comprising: The deviation is determined by the Euclidean distance between the sample point at the current time and the cluster center of the cluster where the sample point is located whether it is greater than the average Euclidean distance between the remaining sample points in the cluster and the cluster center Determined: when greater than , the determination device enters a sub-healthy state and controls the warning.

6. A control and management early warning system for forging production, characterized in that, The method comprises: a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, a method for controlling and early warning of a forging production according to any one of claims 1-5 is realized.

Images