Plastic manufacturing process anomaly detection method and system based on sensor data fusion

Abstract

The application provides a plastic manufacturing process anomaly detection method and system based on sensor data fusion, and relates to the technical field of production process data analysis, and comprises the following steps: arranging multiple sensors on a plastic manufacturing production line; periodically collecting multiple real-time detection values based on the multiple sensors respectively; periodically obtaining sampling detection results of plastic finished products; performing first data fusion analysis based on the current real-time detection values and the sampling detection results to determine whether there is equipment anomaly in the plastic manufacturing production line; if the determination result is that there is equipment anomaly, forming a real-time detection value array and a sampling detection result array in chronological order; and performing second data fusion analysis based on the real-time detection value array and the sampling detection result array. The application has the advantages of overcoming instantaneous data disturbance and realizing high-reliability identification and traceability positioning of abnormal states.

Description

Technical Field

[0001] This invention relates to the field of production process data analysis technology, and more specifically, to a method and system for detecting anomalies in plastic manufacturing processes based on sensor data fusion. Background Technology

[0002] In order to ensure product performance and consistency, anomaly monitoring is usually carried out at various stages of the production line during the plastic manufacturing process.

[0003] Multiple sensors are typically installed in production lines to collect relevant process parameters for anomaly detection. However, current anomaly detection technologies mainly rely on human experience or rule-based threshold judgments. Based on existing technologies, anomaly detection that relies on sensor data is easily affected by environmental fluctuations, noise, drift, or short-term disturbances, causing instantaneous data disturbances. This leads to a risk of missed detections, resulting in frequent false alarms or missed alarms, making it difficult to achieve early warning and accurate location of equipment anomalies.

[0004] Therefore, there is an urgent need to develop methods for detecting anomalies in the plastic manufacturing process, overcome instantaneous data disturbances, and achieve highly reliable identification and traceability of abnormal states, thereby improving the intelligence and stability of detection in the plastic manufacturing process. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting anomalies in plastic manufacturing processes based on sensor data fusion, which can overcome instantaneous data disturbances and thus achieve highly reliable identification and source tracing of abnormal states.

[0006] This invention is achieved through the following technical solution:

[0007] A method for detecting anomalies in plastic manufacturing processes based on sensor data fusion includes the following steps:

[0008] Multiple sensors are deployed on the plastic manufacturing production line;

[0009] Multiple real-time detection values ​​are periodically collected based on various sensors;

[0010] Periodically obtain sampling and testing results for finished plastic products;

[0011] Based on the current real-time detection values ​​and the sampling detection results, a first data fusion analysis is performed. The first data fusion analysis is used to eliminate the influence of detection error by fusing the data of the non-conforming sample density evaluation parameters and the real-time detection values, and to determine whether there is any equipment abnormality in the plastic manufacturing production line.

[0012] If the assessment result indicates that there is no equipment malfunction, no action is taken.

[0013] If the judgment result indicates that there is a device malfunction, then the real-time detection values ​​and the sampling detection results for the N detection cycles up to the current time are obtained, and the real-time detection value array and the sampling detection result array are formed in time sequence respectively.

[0014] A second data fusion analysis is performed based on the real-time detection value array and the sampling detection result array to locate the abnormal parts.

[0015] Preferably, the method for arranging multiple sensors on a plastic manufacturing production line is as follows:

[0016] A temperature sensor is placed in the material heating zone, the inner cavity of the plastic manufacturing mold, and the cooling components of the equipment;

[0017] A pressure sensor is placed at the front end of the material injection component;

[0018] Current sensors and voltage sensors are placed on the circuit connection lines of the manufacturing equipment.

[0019] Preferably, the method for performing the first data fusion analysis based on the real-time detection values ​​and the sampling detection results is as follows:

[0020] Based on the sampling and testing results, non-conforming sample density evaluation parameters are obtained. These non-conforming sample density evaluation parameters are used to comprehensively describe the proportion of non-conforming samples and the degree of parameter deviation in the sampling and testing results.

[0021] The first data fusion analysis is performed by combining the density evaluation parameters of the non-compliant samples and the real-time detection values.

[0022] Preferably, the method for obtaining the density evaluation parameter of the non-compliant sample based on the sampling and testing results is as follows:

[0023] A total of J performance parameters are measured for each sample. If any performance parameter of a sample exceeds the corresponding threshold range, it is judged as an unqualified sample.

[0024] Obtain the percentage of non-compliant samples from the sampling inspection results. :

[0025] ;

[0026] in, This refers to the total number of samples tested. This refers to the number of non-compliant samples detected during sampling inspection;

[0027] The degree of parameter deviation is obtained based on the extent to which the performance parameters of the sampled test samples deviate from the corresponding threshold range. :

[0028] ;

[0029] ;

[0030] ;

[0031] in, The offset value of the j-th performance parameter of the m-th sample. The penalty function value, For the threshold range of the j-th performance parameter, Let be the value of the j-th performance parameter of the m-th sample, and if represent the condition in the conditional function. It is a natural exponential function. The offset tolerance is the value of which is within the range of 1. ;

[0032] The non-compliance sample density evaluation parameter is obtained by combining the proportion of non-compliant samples from the sampling inspection results and the degree of parameter deviation. :

[0033] ;

[0034] in, This is the offset weight adjustment factor. It is a non-linear amplification index. and Used to control the degree of parameter offset The degree of influence on the density evaluation parameters of the non-compliant samples was obtained through fitting training.

[0035] Preferably, the method for the first data fusion analysis is as follows:

[0036] A fusion evaluation function is established based on the density evaluation parameters of the non-conforming samples and the number of outliers in the real-time detection values. The fusion score value R is obtained through the fusion evaluation function.

[0037] ;

[0038] in, As an adjustment constant, This represents the value of the real-time detection data from the s-th sensor. This represents the numerical threshold range of the s-th sensor. Represents the number of sensors. It is a truth-valued function;

[0039] If the value of the fusion score R is greater than the preset score threshold, it is determined that there is a device malfunction; otherwise, it is determined that there is no device malfunction.

[0040] Preferably, the method for performing the second data fusion analysis is as follows:

[0041] From the sampling test result array, all samples with qualified sampling test results are removed, and a sampling non-qualified array is formed in sequence.

[0042] Correlation analysis with adjustment coefficients is performed on the real-time detection value array and the sampled non-conforming array to locate the sensor with abnormal values;

[0043] Perform anomaly analysis on sensors that show abnormal values ​​to determine whether the corresponding points are abnormal;

[0044] Feedback is provided based on the results of the second data fusion analysis, and the adjustment coefficient of the correlation analysis with the adjustment coefficient is updated.

[0045] Preferably, the method for performing correlation analysis with adjustment coefficients based on the real-time detection value array and the sampled non-conforming array is as follows:

[0046] Obtain the types of performance parameters corresponding to all unqualified performance parameters appearing in the unqualified array, and denote them as unqualified performance parameter types;

[0047] For each type of non-conforming performance parameter, obtain the average value for each detection cycle in the non-conforming array. , This represents the average value of the u-th type of non-conforming performance parameter in the non-conforming array during the v-th detection cycle. , , The total number of types of unqualified performance parameters;

[0048] Correlation analysis is performed based on the average value of the u-th type of non-conforming performance parameter in each detection cycle and the value of the s-th sensor in the real-time detection value array, with an adjustment coefficient. , , The method for correlation analysis using the adjustment coefficient, representing the number of sensors, is as follows:

[0049] ;

[0050] ;

[0051] ;

[0052] in, The evaluation parameters for the correlation analysis with the assigned adjustment coefficient, and For array intermediate parameters, A function to determine the correlation between two arrays. This represents the real-time detection value of the s-th sensor in the nth detection cycle, where n = 1, 2, ..., N. The initial value of the adjustment coefficient without any feedback adjustment is 1;

[0053] The evaluation parameters of the correlation analysis with the adjustment coefficient. Greater than the preset evaluation parameter threshold Then the s-th sensor is determined to be the sensor with an abnormal value.

[0054] Preferably, the method for anomaly analysis of sensors exhibiting abnormal values ​​involves performing the following operations for each sensor identified as exhibiting an abnormal value:

[0055] Let the sensor identified as having an abnormal value be the k-th sensor. ;

[0056] The evaluation parameters for the correlation analysis of all the modulated coefficients of the k-th sensor are obtained. , ;

[0057] The anomaly score of the k-th sensor is obtained based on the evaluation parameters obtained from the correlation analysis using the adjusted coefficient. :

[0058] ;

[0059] ;

[0060] in, This represents a function that finds the maximum value based on the change in u. This is the magnification factor. and These are the average value and standard deviation of the k-th sensor in the real-time detection numerical array, respectively, and if(.) is the truth function;

[0061] like The value is greater than the preset anomaly scoring threshold. If the target detected by the k-th sensor is an abnormal part, then it is determined that the target is an abnormal part.

[0062] Preferably, the method for adjusting the weights of the correlation analysis with adjustment coefficients based on the results of the second data fusion analysis is as follows:

[0063] A sensor influence trajectory diagram is established, which includes multiple sensor nodes and multiple performance parameter nodes. The s-th sensor node represents the s-th sensor, and the j-th performance parameter node represents the j-th performance parameter of the plastic. , J represents the number of performance parameters;

[0064] Each time an abnormal location is detected during the positioning process, if the result shows that the target detected by the k-th sensor is an abnormal location, a vector pointing from the k-th sensor node to the d-th performance parameter node is added, and the vector is assigned a weight of 1. The d-th performance parameter is a performance parameter related to the abnormal value of the k-th sensor in the first data fusion analysis. , ;

[0065] After performing the second data fusion analysis, the detection part of the k-th sensor is inspected. If the improvement of the d-th performance parameter after inspection is less than the threshold, the weight of the vector pointing from the k-th sensor node to the d-th performance parameter node is modified. , Otherwise, no action will be taken;

[0066] When performing the correlation analysis with the adjustment coefficient, if the correlation analysis is performed based on the average value of the f-th type of non-conforming performance parameter in each detection cycle and the value of the k-th sensor in the real-time detection value array, the performance parameter node and the k-th sensor node corresponding to the f-th type of non-conforming performance parameter are obtained, and the weights of all vectors between the nodes are obtained. ;

[0067] Check the sensor influence trajectory diagram and update the corresponding adjustment coefficient. :

[0068] ;

[0069] in, To avoid positive numbers with a denominator of 0, Let y be the weight of the y-th vector between nodes, and Y be the total number of vectors between nodes.

[0070] This invention also provides an anomaly detection system for plastic manufacturing processes based on sensor data fusion, applied to the aforementioned anomaly detection method for plastic manufacturing processes based on sensor data fusion, characterized in that it includes:

[0071] Sensor modules are used to deploy various sensors on plastic manufacturing production lines;

[0072] The data detection and acquisition module is used to periodically collect various real-time detection values ​​based on multiple sensors.

[0073] The sampling data acquisition module is used to periodically acquire the sampling and testing results of finished plastic products;

[0074] The first data fusion analysis module is used to perform a first data fusion analysis based on the current real-time detection values ​​and the sampling detection results. The first data fusion analysis is used to eliminate the influence of detection error by fusion of the non-conforming sample density evaluation parameters and the real-time detection values, and to determine whether there is equipment abnormality in the plastic manufacturing production line.

[0075] The time-series array acquisition module is used to acquire the real-time detection values ​​and the sampling detection results up to the current N detection cycles when the judgment result is that there is a device abnormality, and form the real-time detection value array and the sampling detection result array respectively according to the time sequence.

[0076] The second data fusion analysis module is used to perform a second data fusion analysis based on the real-time detection value array and the sampling detection result array to locate the abnormal parts.

[0077] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0078] This invention combines real-time detection values ​​from multiple sensors during the plastic production process with sampling inspection results to achieve integrated analysis of real-time detection and finished product quality inspection. This effectively suppresses inaccurate judgments caused by instantaneous data errors or fluctuations under a single data source, and improves the accuracy and robustness of anomaly detection.

[0079] In the first data fusion, this invention comprehensively considers the proportion of non-conforming samples and the degree of deviation of various performance parameters in the sampling inspection, as well as the sensor values ​​of various key points on the production line, to construct a more comprehensive measurement index of the plastic production status. This index can more comprehensively reflect the process stability and production status, and improve the sensitivity and accuracy of process anomalies.

[0080] In the first data fusion, the present invention determines whether there is an anomaly by using data from a single point in time. It can perform high-speed analysis based on a simple amount of data and respond quickly at the time point where an anomaly may exist. Then, it conducts a more in-depth analysis in the second data fusion based on multi-period data for the extracted suspected anomaly points, realizes the tracing and location of the abnormal parts, and further improves the reliability of anomaly detection.

[0081] This invention is rationally designed, easy to deploy, and applicable to various plastic product manufacturing scenarios, possessing good engineering adaptability and promotional value. Attached Figure Description

[0082] Figure 1 This is a flowchart illustrating the anomaly detection method for plastic manufacturing process based on sensor data fusion provided in Embodiment 1 of the present invention.

[0083] Figure 2A schematic diagram of the principle of the anomaly detection system for plastic manufacturing process based on sensor data fusion provided in Embodiment 2 of the present invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0085] Example 1

[0086] This embodiment provides a method for anomaly detection in plastic manufacturing processes based on sensor data fusion. (See attached document.) Figure 1 This includes the following steps:

[0087] Step S01: Deploy various sensors on the plastic manufacturing production line;

[0088] In this embodiment, the method for arranging multiple sensors on a plastic manufacturing production line is as follows:

[0089] A temperature sensor is placed in the material heating zone, the inner cavity of the plastic manufacturing mold, and the cooling components of the equipment;

[0090] A pressure sensor is placed at the front end of the material injection component;

[0091] Current sensors and voltage sensors are placed on the circuit connection lines of the manufacturing equipment.

[0092] Step S02: Periodically collect various real-time detection values ​​based on multiple sensors, and periodically obtain sampling inspection results of finished plastic products.

[0093] It should be noted that when periodically obtaining sampling test results for finished plastic products, the samples taken for testing are finished plastic products from the previous testing cycle to the current testing cycle.

[0094] When conducting sampling inspections, evaluations can be based on various performance parameters, such as dimensional accuracy, tensile strength, surface roughness detected by a profilometer, and color consistency detected by a colorimeter.

[0095] Step S03: Based on the current real-time detection values ​​and the sampling detection results, a first data fusion analysis is performed. The first data fusion analysis is used to eliminate the influence of detection error by fusing the data of the non-conforming sample density evaluation parameters and the real-time detection values, and to determine whether there is any equipment abnormality in the plastic manufacturing production line.

[0096] As a preferred embodiment, the method for performing the first data fusion analysis based on the real-time detection values ​​and the sampling detection results is as follows:

[0097] Based on the sampling and testing results, non-conforming sample density evaluation parameters are obtained. These non-conforming sample density evaluation parameters are used to comprehensively describe the proportion of non-conforming samples and the degree of parameter deviation in the sampling and testing results.

[0098] The first data fusion analysis is performed by combining the density evaluation parameters of the non-compliant samples and the real-time detection values.

[0099] Specifically, the method for obtaining the density evaluation parameters of unqualified samples based on the sampling and testing results is as follows:

[0100] A total of J performance parameters are measured for each sample. If any performance parameter of a sample exceeds the corresponding threshold range, it is judged as an unqualified sample.

[0101] Obtain the percentage of non-compliant samples from the sampling inspection results. :

[0102] ;

[0103] in, This refers to the total number of samples tested. This refers to the number of non-compliant samples detected during sampling inspection;

[0104] The degree of parameter deviation is obtained based on the extent to which the performance parameters of the sampled test samples deviate from the corresponding threshold range. :

[0105] ;

[0106] ;

[0107] ;

[0108] in, The offset value of the j-th performance parameter of the m-th sample. The penalty function value, For the threshold range of the j-th performance parameter, Let be the value of the j-th performance parameter of the m-th sample, and if represent the condition in the conditional function. It is a natural exponential function, based on experience The offset tolerance is the value of which is within the range of 1. ;

[0109] The non-compliance sample density evaluation parameter is obtained by combining the proportion of non-compliant samples from the sampling inspection results and the degree of parameter deviation. :

[0110] ;

[0111] in, This is the offset weight adjustment factor. It is a non-linear amplification index. and Used to control the degree of parameter offset The degree of influence on the density evaluation parameters of the non-compliant samples was obtained through fitting training.

[0112] Based on this, the method for the first data fusion analysis is as follows:

[0113] A fusion evaluation function is established based on the density evaluation parameters of the non-conforming samples and the number of outliers in the real-time detection values. The fusion score value R is obtained through the fusion evaluation function.

[0114] ;

[0115] in, As an adjustment constant, This represents the value of the real-time detection data from the s-th sensor. This represents the numerical threshold range of the s-th sensor. Represents the number of sensors. It is a truth-valued function;

[0116] If the value of the fusion score R is greater than the preset score threshold, it is determined that there is a device malfunction; otherwise, it is determined that there is no device malfunction.

[0117] In the above scheme, the density of non-conforming samples is used as an evaluation parameter. To quantify the quality level of sampling and detection, the degree of parameter offset was obtained. and the percentage of unqualified samples The parameters are jointly constructed as a comprehensive index for evaluating the density of non-compliant samples. By analyzing data, not limited to sensors, we can eliminate abnormal data analysis biases caused by short-term sensor errors due to external influences. This is applied to the calculation of density evaluation parameters for non-conforming samples. At that time, through A nonlinear amplification mechanism is introduced, which significantly amplifies severe offsets, thereby improving the sensitivity to systemic severe anomalies. An adjustable offset weight adjustment factor is also introduced. If the site requires particular sensitivity to parameter deviations, the [value] can be increased. Through adjustable and trainable and This gives the entire system good flexibility and adaptability. Evaluation parameters for the density of non-conforming samples. The larger the value, the worse the overall quality of the sampling inspection.

[0118] Meanwhile, this embodiment calculates the degree of parameter offset. When, a penalty function was used. By introducing the degree of deviation of each performance parameter from the threshold into the penalty function, the ability to detect slight abnormal fluctuations is significantly enhanced, and the sensitivity of early fault identification is improved by data fusion of multiple performance parameters. This represents the offset value of the j-th performance parameter of the m-th sample, which is the relative deviation of the performance index from its allowable threshold range. Then, for... A penalty function was applied, and the penalty function was applied. Item, in When the value is very small, the penalty function approaches 0; conversely, when the penalty function is large, its value approaches 1. Ultimately, the degree of parameter offset... It represents the average deviation of all non-compliant samples across all testing parameters.

[0119] Because the equipment operation in the plastic manufacturing process has complex nonlinear characteristics, real-time sensor data may fluctuate, and relying solely on real-time data is prone to misjudgment. Furthermore, abnormalities on the production line will inevitably lead to abnormalities in the quality of finished plastic products. Therefore, this implementation calculates the density evaluation parameters for non-conforming samples. Subsequently, anomaly detection is made by combining the quality of the sampling inspection with the detection results from multiple sensors. In other words, this implementation effectively compensates for the lag in sampling inspection and the volatility of sensor data through a data fusion mechanism, resulting in higher robustness and anti-interference capabilities. Specifically, in obtaining the fusion score value R using the fusion evaluation function, the density evaluation parameter of the non-conforming samples is incorporated. The evaluation is based on the number of sensors that are outside the standard threshold range; the more abnormal sensors there are, the better. The larger the value, the better. The processing makes the score growth of this item smoother and also facilitates the adjustment of the sensor's influence. Among them, the adjustment constant... The purpose of v is to prevent the system from reporting errors when ln0 (i.e. all sensors are normal). A larger v will make the parameters inside ln(.) more sensitive. The value of v can also be obtained through fitting experiments.

[0120] If the above methods determine that there is no equipment malfunction, no action is taken, and the process will wait for the next testing cycle.

[0121] If the above solution indicates that there is a device malfunction, proceed to step S04.

[0122] Step S04: Obtain the real-time detection values ​​and the sampling detection results for the N detection cycles up to the current time, and form a real-time detection value array and a sampling detection result array in time sequence respectively;

[0123] Step S05: Perform a second data fusion analysis based on the real-time detection value array and the sampling detection result array to locate the abnormal parts.

[0124] As a preferred embodiment, the method for performing the second data fusion analysis is as follows:

[0125] From the sampling test result array, all samples with qualified sampling test results are removed, and a sampling non-qualified array is formed in sequence; the purpose of this step is to retain only the problematic samples that are helpful in locating anomalies.

[0126] Correlation analysis with adjustment coefficients is performed on the real-time detection value array and the sampled non-conforming array to locate the sensor with abnormal values;

[0127] Perform anomaly analysis on sensors that show abnormal values ​​to determine whether the corresponding points are abnormal;

[0128] Feedback is provided based on the results of the second data fusion analysis, and the adjustment coefficient of the correlation analysis with the adjustment coefficient is updated.

[0129] The above scheme aims to perform multi-point data analysis on sensors identified as abnormal after the initial data fusion analysis, achieving a more accurate further judgment to determine whether an anomaly exists and locate the abnormal location. Furthermore, the scheme adjusts key parameters for locating the anomaly location based on the feedback results, ensuring the reliability of the calculation through real-time updates. In other words, the long-term accuracy is guaranteed after the fusion feedback adjustment mechanism.

[0130] When locating sensors exhibiting abnormal values, the method for performing correlation analysis based on the real-time detected value array and the sampled non-conforming array with adjustment coefficients is as follows:

[0131] Obtain the types of performance parameters corresponding to all unqualified performance parameters appearing in the unqualified array, and denote them as unqualified performance parameter types;

[0132] For each type of non-conforming performance parameter, obtain the average value for each detection cycle in the non-conforming array. , This represents the average value of the u-th type of non-conforming performance parameter in the non-conforming array during the v-th detection cycle. , , The total number of types of unqualified performance parameters;

[0133] Correlation analysis is performed based on the average value of the u-th type of non-conforming performance parameter in each detection cycle and the value of the s-th sensor in the real-time detection value array, with an adjustment coefficient. , , The method for correlation analysis using the adjustment coefficient, representing the number of sensors, is as follows:

[0134] ;

[0135] ;

[0136] ;

[0137] in, The evaluation parameters for the correlation analysis with the assigned adjustment coefficient, and For array intermediate parameters, A function to determine the correlation between two arrays. This represents the real-time detection value of the s-th sensor in the nth detection cycle, where n = 1, 2, ..., N. The initial value of the adjustment coefficient without any feedback adjustment is 1;

[0138] The evaluation parameters of the correlation analysis with the adjustment coefficient. Greater than the preset evaluation parameter threshold Then the s-th sensor is determined to be the sensor with an abnormal value.

[0139] Next, the method for anomaly analysis of sensors exhibiting abnormal values ​​is as follows: for each sensor identified as exhibiting an abnormal value, the following operations are performed:

[0140] Let the sensor identified as having an abnormal value be the k-th sensor. ;

[0141] The evaluation parameters for the correlation analysis of all the modulated coefficients of the k-th sensor are obtained. , ;

[0142] The anomaly score of the k-th sensor is obtained based on the evaluation parameters obtained from the correlation analysis using the adjusted coefficient. :

[0143] ;

[0144] ;

[0145] in, This represents a function that finds the maximum value based on the change in u. This is the magnification factor. and These are the average value and standard deviation of the k-th sensor in the real-time detection numerical array, respectively, and if(.) is the truth function;

[0146] like The value is greater than the preset anomaly scoring threshold. If the target detected by the k-th sensor is an abnormal part, then it is determined that the target is an abnormal part.

[0147] Furthermore, the preferred method for adjusting the weights of the correlation analysis with adjustment coefficients based on the results of the second data fusion analysis is as follows:

[0148] A sensor influence trajectory diagram is established, which includes multiple sensor nodes and multiple performance parameter nodes. The s-th sensor node represents the s-th sensor, and the j-th performance parameter node represents the j-th performance parameter of the plastic. , J represents the number of performance parameters;

[0149] Each time an abnormal location is detected during the positioning process, if the result shows that the target detected by the k-th sensor is an abnormal location, a vector pointing from the k-th sensor node to the d-th performance parameter node is added, and the vector is assigned a weight of 1. The d-th performance parameter is a performance parameter related to the abnormal value of the k-th sensor in the first data fusion analysis. , ;

[0150] After performing the second data fusion analysis, the detection part of the k-th sensor is inspected. If the improvement of the d-th performance parameter after inspection is less than the threshold, the weight of the vector pointing from the k-th sensor node to the d-th performance parameter node is modified. , Otherwise, no action will be taken;

[0151] When performing the correlation analysis with the adjustment coefficient, if the correlation analysis is performed based on the average value of the f-th type of non-conforming performance parameter in each detection cycle and the value of the k-th sensor in the real-time detection value array, the performance parameter node and the k-th sensor node corresponding to the f-th type of non-conforming performance parameter are obtained, and the weights of all vectors between the nodes are obtained. ;

[0152] Check the sensor influence trajectory diagram and update the corresponding adjustment coefficient. :

[0153] ;

[0154] in, To avoid positive numbers with a denominator of 0, Let y be the weight of the y-th vector between nodes, and Y be the total number of vectors between nodes.

[0155] In the above scheme, the first step is to use correlation analysis with adjustment coefficients to identify abnormal performance parameters. This involves using the type of non-compliant performance parameter as a baseline to deduce the location of the anomaly through inverse data analysis. The correlation analysis with adjustment coefficients uses the correlation of time series data to find sensor values ​​highly correlated with the non-compliant performance parameters, thus locating suspected abnormal sensors. Further analysis of these suspected abnormal sensor values ​​is then conducted to determine if a problem exists. It is worth noting that when performing correlation analysis with adjustment coefficients, aggregating the non-compliant data at the sample level into a time series helps to align the analysis with real-time sensor data. r(.,.) in the equation can be calculated using the Pearson correlation coefficient.

[0156] Next, an anomaly score can be calculated to further confirm the presence of abnormal sensors. Anomaly score The calculation incorporates correlation and sensor variability, directly introducing evaluation parameters from correlation analysis with adjustment coefficients, as well as the ratio of the standard deviation to the mean of the sensors in the real-time detection data array. If abnormal sensor data fluctuates significantly and is highly correlated with unacceptable results, the score naturally increases. This is because sensors may affect multiple performance parameters. To avoid false alarms, the relevant values ​​of the performance parameters with the greatest impact are directly used for judgment. In other words, if the sensor fluctuates significantly and affects any performance parameter, it is judged as an abnormal point.

[0157] The adjustable coefficient in this embodiment supports a subsequent feedback mechanism, improving the method's adaptability. The feedback rule primarily involves lowering the edge weights if performance improvement is minimal after maintenance; this weight change affects the value of the adjustable coefficient in the next round of correlation analysis. Furthermore, a graph model is incorporated to ensure the association weights have historical memory. This feedback mechanism effectively avoids continuous misjudgment due to one or two misjudgments, improving judgment accuracy.

[0158] Example 2

[0159] This invention also provides an anomaly detection system for plastic manufacturing processes based on sensor data fusion, see below. Figure 2 The method for detecting anomalies in the plastic manufacturing process based on sensor data fusion, as described above, is characterized by comprising:

[0160] Sensor modules are used to deploy various sensors on plastic manufacturing production lines;

[0161] The data detection and acquisition module is used to periodically collect various real-time detection values ​​based on multiple sensors.

[0162] The sampling data acquisition module is used to periodically acquire the sampling and testing results of finished plastic products;

[0163] The first data fusion analysis module is used to perform a first data fusion analysis based on the current real-time detection values ​​and the sampling detection results. The first data fusion analysis is used to eliminate the influence of detection error by fusion of the non-conforming sample density evaluation parameters and the real-time detection values, and to determine whether there is equipment abnormality in the plastic manufacturing production line.

[0164] The time-series array acquisition module is used to acquire the real-time detection values ​​and the sampling detection results up to the current N detection cycles when the judgment result is that there is a device abnormality, and form the real-time detection value array and the sampling detection result array respectively according to the time sequence.

[0165] The second data fusion analysis module is used to perform a second data fusion analysis based on the real-time detection value array and the sampling detection result array to locate the abnormal parts.

[0166] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting anomalies in plastic manufacturing processes based on sensor data fusion, characterized in that, Includes the following steps: Multiple sensors are deployed on the plastic manufacturing production line; Multiple real-time detection values ​​are periodically collected based on various sensors; Periodically obtain sampling and testing results for finished plastic products; Based on the current real-time detection values ​​and the sampling detection results, a first data fusion analysis is performed. The first data fusion analysis is used to eliminate the influence of detection error by fusing the data of the non-conforming sample density evaluation parameters and the real-time detection values, and to determine whether there is any equipment abnormality in the plastic manufacturing production line. If the assessment result indicates that there is no equipment malfunction, no action is taken. If the judgment result indicates that there is a device malfunction, then the real-time detection values ​​and the sampling detection results for the N detection cycles up to the current time are obtained, and the real-time detection value array and the sampling detection result array are formed in time sequence respectively. A second data fusion analysis is performed based on the real-time detection value array and the sampling detection result array to locate the abnormal parts.

2. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 1, characterized in that, The method for deploying multiple sensors on a plastic manufacturing production line is as follows: A temperature sensor is placed in the material heating zone, the inner cavity of the plastic manufacturing mold, and the cooling components of the equipment; A pressure sensor is placed at the front end of the material injection component; Current sensors and voltage sensors are placed on the circuit connection lines of the manufacturing equipment.

3. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 1, characterized in that, The method for performing the first data fusion analysis based on the real-time detection values ​​and the sampling detection results is as follows: Based on the sampling and testing results, non-conforming sample density evaluation parameters are obtained. These non-conforming sample density evaluation parameters are used to comprehensively describe the proportion of non-conforming samples and the degree of parameter deviation in the sampling and testing results. The first data fusion analysis is performed by combining the density evaluation parameters of the non-compliant samples and the real-time detection values.

4. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 3, characterized in that, The method for obtaining density evaluation parameters of non-compliant samples based on the sampling and testing results is as follows: A total of J performance parameters are measured for each sample. If any performance parameter of a sample exceeds the corresponding threshold range, it is judged as an unqualified sample. Obtain the percentage of non-compliant samples from the sampling inspection results. : ； in, This refers to the total number of samples tested. This refers to the number of non-compliant samples detected during the sampling inspection. The degree of parameter deviation is obtained based on the extent to which the performance parameters of the sampled test samples deviate from the corresponding threshold range. : ； ； ； in, The offset value of the j-th performance parameter of the m-th sample. The penalty function value, For the threshold range of the j-th performance parameter, Let be the value of the j-th performance parameter of the m-th sample, and if represent the condition in the conditional function. It is a natural exponential function. The offset tolerance is the value that can be set within a certain range. ; The non-compliance sample density evaluation parameter is obtained by combining the proportion of non-compliant samples from the sampling inspection results and the degree of parameter deviation. : ； in, This is the offset weight adjustment factor. It is a non-linear amplification index. and Used to control the degree of parameter offset The degree of influence on the density evaluation parameters of the non-compliant samples was obtained through fitting training.

5. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 4, characterized in that, The method for the first data fusion analysis is as follows: A fusion evaluation function is established based on the density evaluation parameters of the non-conforming samples and the number of outliers in the real-time detection values. The fusion score value R is obtained through the fusion evaluation function. ； in, As an adjustment constant, This represents the value of the real-time detection data from the s-th sensor. This represents the numerical threshold range of the s-th sensor. Represents the number of sensors. It is a truth-valued function; If the value of the fusion score R is greater than the preset score threshold, it is determined that there is a device malfunction; otherwise, it is determined that there is no device malfunction.

6. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 1, characterized in that, The method for performing the second data fusion analysis is as follows: From the sampling test result array, all samples with qualified sampling test results are removed, and a sampling non-qualified array is formed in sequence. Correlation analysis with adjustment coefficients is performed on the real-time detection value array and the sampled non-conforming array to locate the sensor with abnormal values; Perform anomaly analysis on sensors that show abnormal values ​​to determine whether the corresponding points are abnormal; Feedback is provided based on the results of the second data fusion analysis, and the adjustment coefficient of the correlation analysis with the adjustment coefficient is updated.

7. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 6, characterized in that, The method for performing correlation analysis with adjustment coefficients based on the real-time detection numerical array and the sampled non-conforming array is as follows: Obtain the types of performance parameters corresponding to all unqualified performance parameters appearing in the unqualified array, and denote them as unqualified performance parameter types; For each type of non-conforming performance parameter, obtain the average value for each detection cycle in the non-conforming array. , This represents the average value of the u-th type of non-conforming performance parameter in the non-conforming array during the v-th detection cycle. , , The total number of types of unqualified performance parameters; Correlation analysis is performed based on the average value of the u-th type of non-conforming performance parameter in each detection cycle and the value of the s-th sensor in the real-time detection value array, with an adjustment coefficient. , , The method for correlation analysis using the adjustment coefficient, representing the number of sensors, is as follows: ； ； ； in, The evaluation parameters for the correlation analysis with the assigned adjustment coefficient, and For array intermediate parameters, A function to determine the correlation between two arrays. This represents the real-time detection value of the s-th sensor in the nth detection cycle, where n = 1, 2, ..., N. The initial value of the adjustment coefficient without any feedback adjustment is 1; The evaluation parameters of the correlation analysis with the adjustment coefficient. Greater than the preset evaluation parameter threshold Then the s-th sensor is determined to be the sensor with an abnormal value.

8. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 7, characterized in that, The method for anomaly analysis of sensors exhibiting abnormal values ​​involves performing the following operations for each sensor identified as having an abnormal value: Let the sensor identified as having an abnormal value be the k-th sensor. ; The evaluation parameters for the correlation analysis of all the modulated coefficients of the k-th sensor are obtained. , ; The anomaly score of the k-th sensor is obtained based on the evaluation parameters obtained from the correlation analysis using the adjusted coefficient. : ； ； in, This represents a function that finds the maximum value based on the change in u. This is the magnification factor. and These are the average value and standard deviation of the k-th sensor in the real-time detection numerical array, respectively, and if(.) is the truth function; like The value is greater than the preset anomaly scoring threshold. If the target detected by the k-th sensor is an abnormal part, then it is determined that the target is an abnormal part.

9. The method for anomaly detection in plastic manufacturing process based on sensor data fusion according to claim 8, characterized in that, The method for adjusting the weights of the correlation analysis with the adjustment coefficient based on the results of the second data fusion analysis is as follows: A sensor influence trajectory diagram is established, which includes multiple sensor nodes and multiple performance parameter nodes. The s-th sensor node represents the s-th sensor, and the j-th performance parameter node represents the j-th performance parameter of the plastic. , J represents the number of performance parameters; Each time an abnormal location is detected during the positioning process, if the result shows that the target detected by the k-th sensor is an abnormal location, a vector pointing from the k-th sensor node to the d-th performance parameter node is added, and the vector is assigned a weight of 1. The d-th performance parameter is a performance parameter related to the abnormal value of the k-th sensor in the first data fusion analysis. , ; After performing the second data fusion analysis, the detection part of the k-th sensor is inspected. If the improvement of the d-th performance parameter after inspection is less than the threshold, the weight of the vector pointing from the k-th sensor node to the d-th performance parameter node is modified. , Otherwise, no action will be taken; When performing the correlation analysis with the adjustment coefficient, if the correlation analysis is performed based on the average value of the f-th type of non-conforming performance parameter in each detection cycle and the value of the k-th sensor in the real-time detection value array, the performance parameter node and the k-th sensor node corresponding to the f-th type of non-conforming performance parameter are obtained, and the weights of all vectors between the nodes are obtained. ; Check the sensor influence trajectory diagram and update the corresponding adjustment coefficient. : ； in, To avoid positive numbers with a denominator of 0, Let y be the weight of the y-th vector between nodes, and Y be the total number of vectors between nodes.

10. A plastic manufacturing process anomaly detection system based on sensor data fusion, applied to the plastic manufacturing process anomaly detection method based on sensor data fusion as described in any one of claims 1-9, characterized in that, include: Sensor modules are used to deploy various sensors on plastic manufacturing production lines; The data detection and acquisition module is used to periodically collect various real-time detection values ​​based on multiple sensors. The sampling data acquisition module is used to periodically acquire the sampling and testing results of finished plastic products; The first data fusion analysis module is used to perform a first data fusion analysis based on the current real-time detection values ​​and the sampling detection results. The first data fusion analysis is used to eliminate the influence of detection error by fusion of the non-conforming sample density evaluation parameters and the real-time detection values, and to determine whether there is equipment abnormality in the plastic manufacturing production line. The time-series array acquisition module is used to acquire the real-time detection values ​​and the sampling detection results up to the current N detection cycles when the judgment result is that there is a device abnormality, and form the real-time detection value array and the sampling detection result array respectively according to the time sequence. The second data fusion analysis module is used to perform a second data fusion analysis based on the real-time detection value array and the sampling detection result array to locate the abnormal parts.

Images