A Method and System for Health Status Assessment of Packaging Machinery Based on Multi-Source Data Fusion

Abstract

This invention discloses a method and system for assessing the health status of packaging machinery based on multi-source data fusion, belonging to the field of multi-source data fusion technology. It includes locating the root cause of faults based on the feature vectors obtained from the decomposition of the principal eigenvalues, thus identifying the main assessment dimensions leading to a decline in the overall health index. This invention achieves precise location of the root cause of faults by analyzing the assessment dimensions mapped to the component with the largest absolute value in the feature vector corresponding to the principal eigenvalue. This process, based on multivariate statistical principles, effectively isolates the coupling correlation between data from different dimensions, avoiding the misjudgments and omissions caused by the reliance on the experience of maintenance personnel in traditional technologies. When the overall health index of the equipment declines, it can quickly pinpoint whether the problem originates from electrical, mechanical, process, or environmental dimensions, significantly shortening fault diagnosis time, improving maintenance response efficiency, and providing a clear direction for targeted equipment repair.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion technology, and in particular to a method and system for assessing the health status of packaging machinery based on multi-source data fusion. Background Technology

[0002] As the packaging industry transforms towards high-speed and automated production, packaging machinery, as core equipment in production lines for food, pharmaceuticals, and daily chemicals, directly determines production efficiency, product quality, and cost through its continuous and stable operation. Long-term high-load operation, coupled with environmental dust, temperature fluctuations, and process switching, makes equipment susceptible to problems such as electrical aging, mechanical wear, and process parameter drift. Sudden failures not only lead to production line downtime losses but may also trigger product recalls. Therefore, real-time and accurate assessment and early warning of its health status have become a key requirement for enterprises to ensure production continuity and reduce operating costs.

[0003] Currently, health assessment of packaging machinery still relies mainly on traditional monitoring and experience-based judgment. Although some areas have introduced intelligent methods, their limitations are obvious: traditional methods mostly depend on single-dimensional data, such as monitoring only mechanical vibration or motor current, and a few combine process parameters for simple analysis, without forming a multi-dimensional collaborative system. In recent years, the development of the Industrial Internet of Things has promoted attempts to integrate multi-source data, but most of these attempts only involve data splicing and independent analysis, without exploring the coupling relationships between various dimensions, such as mechanical abnormalities leading to process fluctuations. Moreover, assessments often use fixed thresholds, lacking a comprehensive quantification of the overall health status, making it difficult to shift from "post-fault repair" to "pre-fault early warning." The root cause location of faults also relies heavily on human experience, resulting in low efficiency.

[0004] Several key issues remain in the current application of this technology: First, the depth of data fusion is insufficient, either neglecting data correlation leading to biased assessments or simply weighting and splicing data without extracting deep health information; second, there is a lack of comprehensive health measurement capabilities, with outputs often being binary, such as normal or abnormal, making it difficult to predict degradation trends and resulting in poor timeliness of early warnings; third, the accuracy of fault root cause localization is low, human experience is easily influenced by subjectivity, and misjudgments are prone to occur when multiple parameters fluctuate, prolonging maintenance time; fourth, the model has poor adaptability, with dimension weights and early warning thresholds mostly being fixed values, unable to be dynamically optimized according to equipment operating cycles or conditions, leading to decreased accuracy after long-term use and making it difficult to meet the needs of refined operation and maintenance.

[0005] Therefore, it is essential to invent a method and system for assessing the health status of packaging machinery based on multi-source data fusion to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for assessing the health status of packaging machinery based on multi-source data fusion, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for assessing the health status of packaging machinery based on multi-source data fusion, comprising the following steps:

[0008] S1. Collect operational data of packaging machinery in multiple preset dimensions, obtain data of at least two monitoring indicators for each dimension, and construct feature vectors corresponding to each dimension;

[0009] The multiple preset dimensions are four evaluation dimensions, including electrical dimension, mechanical dimension, process dimension and environmental dimension;

[0010] The electrical dimension monitoring data includes motor current harmonic distortion rate (THD) and drive power factor (PF), corresponding to the feature vector V. elec =[THD,PF];

[0011] The monitoring data for the mechanical dimension includes the effective value of the vibration velocity V of the drive-end bearing. rms and gearbox vibration signal kurtosis value K urt The corresponding feature vector V mech =[V rms K urt ];

[0012] The monitoring indicators for the process dimension include the yield rate P per unit time. rat and the standard deviation of sealing temperature fluctuation T std The corresponding feature vector V proc =[P rat T std ];

[0013] The environmental monitoring data includes the spindle bearing temperature T. bea and environmental dust concentration D ust The corresponding feature vector V env =[T bea D ust ];

[0014] S2. Combine the feature vectors corresponding to each dimension into a health status matrix, and then normalize it to obtain a normalized matrix.

[0015] The normalization process is implemented using a membership function, which maps the elements of the health status matrix to the interval [0, 1]. Specifically, for monitoring index data where lower values ​​represent better equipment status, a descending semi-trapezoidal membership function is used, while for monitoring index data where higher values ​​represent better equipment status, an ascending semi-trapezoidal membership function is used.

[0016] S3. Assign weights to each dimension, construct a weight diagonal matrix, and multiply the normalized matrix with the weight diagonal matrix to obtain the weighted state matrix;

[0017] S4. Calculate the covariance matrix of the weighted state matrix, perform eigenvalue decomposition to obtain the principal eigenvalue, and then calculate the comprehensive health index based on the principal eigenvalue;

[0018] The calculation formula for the comprehensive health index is:

[0019] HI = 100×(1 - λ1 / λ ideal )

[0020] where HI is the comprehensive health index, λ1 is the current principal eigenvalue, and λ ideal is the maximum principal eigenvalue of the device in a brand-new ideal state;

[0021] S5. Determine the health state of the device based on the comprehensive health index, and issue a warning when the comprehensive health index is lower than the set threshold;

[0022] S6. Locate the root cause of the failure based on the eigenvector obtained from the decomposition of the principal eigenvalue, and obtain the main evaluation dimension that causes the decline of the comprehensive health index;

[0023] S7. Adaptively optimize the weight diagonal matrix and the threshold according to the comparison result between the main evaluation dimension and the maintenance feedback data;

[0024] The specific process of the adaptive optimization is as follows:

[0025] S71. Record the cause of failure R after the maintenance of the industrial device, the monitoring index data before and after the maintenance, and the change in the comprehensive health index HI;

[0026] S72. Compare the consistency between the main evaluation dimension and the evaluation dimension corresponding to the cause of failure R, and count the number of positioning deviations;

[0027] S73. When the number of positioning deviations exceeds the preset threshold, use the gradient descent algorithm to adjust the diagonal elements of the weight diagonal matrix;

[0028] S74. Dynamically update the threshold in step S5 based on the normal fluctuation range of HI in different operating cycles or different environments of the industrial device.

[0029] Preferably, the determination of the health state of the device is specifically:

[0030] When the comprehensive health index HI ≥ a, it is determined to be in a healthy state;

[0031] When the comprehensive health index b ≤ HI < a, it is determined to be in a sub-healthy state;

[0032] When the comprehensive health index c ≤ HI < b, it is determined to be in a warning state;

[0033] When the comprehensive health index HI < c, it is determined as the alarm state.

[0034] Preferably, the fault root cause location is realized by analyzing the dimension corresponding to the component with the largest absolute value in the eigenvector corresponding to the main eigenvalue.

[0035] The packaging machinery health status evaluation system based on multi-source data fusion includes the following modules:

[0036] The data acquisition and feature construction module is used to collect the operation data of the packaging machinery in multiple preset dimensions, obtain at least two monitoring index data for each dimension, and construct the eigenvector corresponding to each dimension;

[0037] The data normalization processing module is used to combine the eigenvectors corresponding to each dimension into a health status matrix and perform normalization processing to obtain a normalized matrix;

[0038] The weight assignment and weighted calculation module is used to assign weights to each dimension, construct a weight diagonal matrix, and multiply the normalized matrix by the weight diagonal matrix to obtain a weighted status matrix;

[0039] The comprehensive health index calculation module is used to calculate the covariance matrix of the weighted status matrix, perform eigenvalue decomposition to obtain the main eigenvalue, and then calculate the comprehensive health index according to the main eigenvalue;

[0040] The health status evaluation and early warning module is used to determine the equipment health status according to the comprehensive health index, and issue an early warning when the comprehensive health index is lower than the set threshold;

[0041] The fault root cause location module is used to perform fault root cause location according to the eigenvector obtained by decomposing the main eigenvalue, and obtain the main evaluation dimension that causes the decline of the comprehensive health index;

[0042] The adaptive optimization module is used to adaptively optimize the weight diagonal matrix and the threshold according to the comparison result between the main evaluation dimension and the maintenance feedback data.

[0043] The technical effects and advantages of the present invention:

[0044] 1. By collecting the operation data of the packaging machinery in four preset dimensions of electricity, machinery, technology, and environment, constructing an eigenvector containing at least two monitoring indexes for each dimension, and combining the eigenvectors of each dimension into a health status matrix for subsequent analysis, the present invention breaks the limitation of the traditional evaluation method that only relies on single-dimensional data, comprehensively covers the key influencing factors of equipment operation, makes the health status evaluation more conform to the actual operation scenario of the equipment, effectively avoids the one-sidedness of evaluation caused by data dimension missing, and lays a data foundation for accurately evaluating the overall health status of the equipment;

[0045] 2. This invention normalizes the health status matrix by using membership functions. For monitoring indicators where lower values ​​represent better equipment status, a descending half-trapezoidal membership function is used, while for monitoring indicators where higher values ​​represent better equipment status, an ascending half-trapezoidal membership function is used. This unifies the mapping of monitoring data with different dimensions and evaluation standards to the [0, 1] interval, completely eliminating the interference of data scale differences on the evaluation results. This provides standardized data support for subsequent weight allocation and weighted calculation of each dimension, significantly improving the accuracy and reliability of the evaluation results.

[0046] 3. This invention achieves precise fault location by analyzing the evaluation dimension mapped by the component with the largest absolute value in the feature vector corresponding to the principal feature value. This process effectively removes the coupling correlation between data of various dimensions based on the principle of multivariate statistics, avoiding the misjudgment and omission caused by the reliance on the experience judgment of operation and maintenance personnel in traditional technology. When the overall health index of the equipment declines, it can quickly identify whether the problem originates from the electrical, mechanical, process, or environmental dimensions, greatly shortening the fault investigation time, improving the maintenance response efficiency, and providing a clear direction for targeted equipment maintenance.

[0047] 4. This invention records the causes of equipment failures after maintenance, monitoring index data before and after maintenance, and changes in the comprehensive health index. It compares the main evaluation dimensions obtained from fault root cause localization with the actual evaluation dimensions corresponding to the fault causes. When the number of localization deviations exceeds a preset threshold, the gradient descent algorithm is used to adjust the weight diagonal matrix. At the same time, the warning threshold is dynamically updated based on the normal fluctuation range of HI under different equipment operating cycles or different environments. This enables the evaluation model to be continuously optimized as the equipment operating status changes and the working conditions are adjusted, effectively avoiding the problem of long-term evaluation accuracy decline caused by fixed parameters and significantly enhancing the adaptability of the model in complex production scenarios.

[0048] 5. This invention constructs a closed-loop equipment health management system of "monitoring-assessment-early warning-location-optimization" through a full-process design including data collection and feature construction, data normalization processing, weight allocation and weighted calculation, comprehensive health index calculation, health status assessment and early warning, fault root cause location, and adaptive optimization. It can not only provide real-time early warning of potential equipment faults and avoid production line downtime losses and product quality risks caused by sudden failures, but also reduce the problems of over-maintenance or under-maintenance through accurate root cause location. Combined with adaptive optimization of the model, it can further reduce the enterprise's operation and maintenance costs, and ultimately ensure the continuous and stable operation of the packaging production line and the consistency of product quality. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0050] Figure 2This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0051] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] This invention provides, for example Figure 1 The packaging machinery health status assessment method based on multi-source data fusion shown includes the following steps:

[0053] S1. Collect operational data of packaging machinery in multiple preset dimensions, obtain data of at least two monitoring indicators for each dimension, and construct feature vectors corresponding to each dimension;

[0054] It should be noted that the operational data includes electrical data, mechanical data, process data, and environmental data;

[0055] Electrical data is acquired in real time by using a power quality analyzer or smart meter installed in the power supply circuit of the main drive motor of the equipment to collect the current and voltage waveform signals of the motor.

[0056] Mechanical data is acquired by vibration acceleration sensors installed on the drive end bearing housing and gearbox housing to collect mechanical vibration signals, which are then converted from analog to digital by a data acquisition card to obtain vibration time-domain signals.

[0057] Process data is collected by temperature sensors installed on the sealing device to acquire temperature signals; the yield rate data is obtained by the visual inspection system or product counter at the end of the production line and read through the interface of the equipment monitoring system (such as SCADA system);

[0058] Environmental data is collected by temperature sensors installed at measurement points near the spindle bearing to acquire temperature signals; and by dust concentration sensors installed in critical areas of the equipment (such as near the electrical control cabinet) to acquire environmental dust data.

[0059] Furthermore, in the above technical solution, the multiple preset dimensions are four evaluation dimensions, including electrical dimension, mechanical dimension, process dimension and environmental dimension.

[0060] Furthermore, in the above technical solution, the monitoring indicators of the electrical dimension include the motor current harmonic distortion rate (THD) and the drive power factor (PF), corresponding to the feature vector V. elec =[THD,PF];

[0061] The monitoring data for the mechanical dimension includes the effective value of the vibration velocity V of the drive-end bearing. rms and gearbox vibration signal kurtosis value K urt The corresponding feature vector V mech =[V rms K urt ];

[0062] The monitoring indicators for the process dimension include the yield rate P per unit time. rat and the standard deviation of sealing temperature fluctuation T std The corresponding feature vector V proc =[P rat T std ];

[0063] The environmental monitoring data includes the spindle bearing temperature T. bea and environmental dust concentration D ust The corresponding feature vector V env =[T bea D ust ].

[0064] It should be noted that the motor current harmonic distortion rate (THD) is directly output by the power quality analyzer after performing Fourier transform analysis on the collected current waveform, or it is calculated by the industrial control computer based on the collected waveform data. It is used to quantify the degree of distortion of the motor input current waveform relative to the standard sine wave. An abnormal increase in its value often indicates that there is a potential fault in the motor or driver or an abnormal load.

[0065] The driving power factor (PF) is directly measured and output by a power quality analyzer or smart meter, reflecting the efficiency of power utilization. A low power factor indicates reactive power loss, which may be related to motor aging, load mismatch, or power supply quality problems.

[0066] The effective value of the vibration velocity V of the drive end bearing rms The vibration intensity of a bearing is obtained by integrating and calculating the root mean square of the time-domain signal collected by the vibration acceleration sensor. It is a commonly used parameter for evaluating the overall vibration intensity of the bearing and is used to judge bearing wear, loosening or imbalance.

[0067] Transmission vibration signal kurtosis value K urt The value of the time-domain signal acquired by the vibration acceleration sensor is obtained by calculating the higher-order statistics (fourth-order moment). It is very sensitive to the impact component in the signal. Its increase usually indicates that there is early pitting, cracks and other local damage in the gearbox gears, bearings and other components.

[0068] The pass rate P per unit time rat The total output and qualified output within the statistical period of the production line monitoring system are calculated using the formula P. rat=Qualified output / Total output, its value directly reflects the production efficiency and output quality of the equipment, and is a comprehensive indicator for measuring the process performance of the equipment;

[0069] The standard deviation of the sealing temperature fluctuation T std The standard deviation of the continuous temperature values ​​collected by the temperature sensor within a monitoring cycle is calculated to quantify the stability of temperature control in the sealing process. Excessive fluctuations directly affect the sealing quality and may be caused by aging of the heating element or failure of the temperature control system.

[0070] The spindle bearing temperature T bea The temperature is directly measured by a PT100 resistance temperature detector or thermocouple temperature sensor installed on the bearing housing. It monitors the bearing operating temperature. Abnormal temperature rise is an important sign of poor lubrication, accelerated wear, or overload.

[0071] The environmental dust concentration D ust The dust level in the surrounding environment of the equipment is measured and output by a laser dust sensor or particle counter. If the value is too high, the dust will aggravate mechanical wear, affect heat dissipation, and even cause electrical failures.

[0072] S2. Combine the feature vectors corresponding to each dimension into a health status matrix, and then normalize it to obtain a normalized matrix.

[0073] It's important to know that the feature vectors of each evaluation dimension are treated as rows of a matrix and combined in a predetermined order to form a multi-row, two-column matrix called the health status matrix X. For example, the data collected at time t, after being constructed, yields feature vectors for the four dimensions as follows:

[0074] Electrical dimension feature vector: V elec (t)=[THD(t),PF(t)];

[0075] Mechanical dimension feature vector: V mech (t)=[V rms (t), K urt (t)];

[0076] Process dimension feature vector: V proc (t)=[P rat (t), T std (t)];

[0077] Environmental dimension feature vector: V env (t)=[T bea (t), D ust (t)];

[0078] The health state matrix X(t) at this moment is constructed as follows:

[0079] ,

[0080] The matrix X(t) is a 4×2 matrix, where each row represents an evaluation dimension and each column represents the data sequence of a specific monitoring indicator under that dimension at different times.

[0081] Furthermore, in the above technical solution, the normalization process is implemented using a membership function, which maps the elements of the health status matrix to the interval [0, 1]. Specifically, for monitoring index data where lower values ​​represent better equipment status, a descending half-trapezoidal membership function is used, and for monitoring index data where higher values ​​represent better equipment status, an ascending half-trapezoidal membership function is used.

[0082] It is necessary to know the membership function type for each monitoring indicator:

[0083] For the electrical dimension: the motor current harmonic distortion rate (THD) adopts a decreasing half-trapezoidal membership function; the drive power factor (PF) adopts an increasing half-trapezoidal membership function.

[0084] For the mechanical dimension: the effective value of the vibration velocity V of the drive-end bearing. rms A semi-trapezoidal membership function is used; the kurtosis value K of the gearbox vibration signal is... urt A descending semi-trapezoidal membership function is used.

[0085] For the process dimension: Qualified product rate P per unit time rat Using a semi-trapezoidal membership function; standard deviation T of sealing temperature fluctuation. std A descending semi-trapezoidal membership function is used.

[0086] Regarding the environmental dimension: spindle bearing temperature T bea Using a reduced semi-trapezoidal membership function; environmental dust concentration D ust A descending semi-trapezoidal membership function is used.

[0087] The descending half-trapezoidal membership function is defined as follows: Let the actual value of the monitoring indicator data be x, and its optimal health status range be [m, n]. Then the normalized value μ(x) is calculated as follows:

[0088] ,

[0089] The ascending semi-trapezoidal membership function is defined as follows: Let the actual value of the monitoring indicator data be y, and its optimal health status range be [g, h]. Then the normalized value μ(y) is calculated as follows:

[0090] ,

[0091] Among them, m, n, g and h are set according to historical equipment data or expert experience.

[0092] S3. Assign weights to each dimension, construct a weight diagonal matrix, and multiply the normalized matrix with the weight diagonal matrix to obtain the weighted state matrix;

[0093] It should be noted that the values ​​of the weights are determined based on historical equipment failure data, domain expert experience, or the analytic hierarchy process, and satisfy the condition that the sum of all weights is 1.

[0094] S4. Calculate the covariance matrix of the weighted state matrix, perform eigenvalue decomposition to obtain the principal eigenvalues, and then calculate the comprehensive health index based on the principal eigenvalues.

[0095] It should be noted that the weighted state matrix is ​​denoted as X. weighted The covariance matrix is ​​denoted as C, and the formula for calculating the covariance matrix of the weighted state matrix is:

[0096] ,

[0097] Where N is the number of samples, i.e. the number of rows in the matrix, which is 4 in this example, and T represents the transpose operation of the matrix. The calculated covariance matrix C is a 2×2 square matrix.

[0098] The eigenvalue decomposition satisfies the following relationship: , where λ i v represents the eigenvalues ​​of the covariance matrix C. i For the corresponding eigenvectors, the solved eigenvalues ​​are arranged in descending order and denoted as λ1, λ2, λ3, and λ4. Among them, λ1 is the largest eigenvalue, called the principal eigenvalue, which represents the variance of the data in the main direction of change and best reflects the overall trend of the equipment's health status.

[0099] Furthermore, in the above technical solution, the formula for calculating the comprehensive health index is:

[0100] HI = 100 × (1 - λ1 / λ) ideal ),

[0101] Where HI is the comprehensive health index, λ1 is the current main feature value, and λ ideal This represents the maximum principal characteristic value of the device under ideal conditions.

[0102] It is important to know that the maximum principal eigenvalue λ of the device in a completely new ideal state ideal The following method is used to determine λ: In the initial stage after the equipment is installed, debugged, and put into operation, multiple sets of normal operation data are continuously collected. Multiple sets of principal characteristic values ​​are calculated and recorded according to steps S1 to S4, and the maximum value among them is taken as λ. ideal Alternatively, the ideal value specified during the equipment's factory acceptance test can be used directly.

[0103] The meaning of the calculation formula for the comprehensive health index is as follows: When the device is in a brand-new state, its main characteristic value λ1 should be close to λ ideal , at this time HI≈100, indicating excellent health. As the device performance degrades, the main characteristic value λ1 will gradually increase, resulting in a decrease in the HI value; when λ1 = λ ideal , HI = 0, indicating extremely poor health status.

[0104] S5. Determine the health status of the device based on the comprehensive health index, and issue a warning when the comprehensive health index is lower than the set threshold;

[0105] Furthermore, in the above technical solution, the determination of the device health status is specifically as follows:

[0106] When the comprehensive health index HI≥a, it is determined to be in a healthy state;

[0107] When the comprehensive health index b≤HI < a, it is determined to be in a sub-healthy state;

[0108] When the comprehensive health index c≤HI < b, it is determined to be in a warning state;

[0109] When the comprehensive health index HI < c, it is determined to be in an alarm state.

[0110] It should be noted that the thresholds a, b, and c are determined according to the device type, operating conditions, and historical operation data. Exemplarily, a = 85, b = 75, and c = 60 can be taken;

[0111] The healthy state indicates that the operating indicators of all dimensions of the device are good, and the overall performance is at an ideal level. It can be safely put into production, and only routine maintenance and condition monitoring are required;

[0112] The sub-healthy state indicates that the device has shown slight performance degradation or individual indicators have slightly deviated from the optimal range. Although it does not affect the current normal operation, it is necessary to strengthen the monitoring frequency, pay attention to its change trend, and can start to prepare for preventive maintenance;

[0113] The warning state indicates that the device performance has significantly degraded, and there is a high risk of failure. The system should immediately issue a warning signal, such as highlighting a yellow alarm on the monitoring interface, sending an in-station message or an email notification, prompting the maintenance personnel to intervene. It is necessary to combine the root cause location result of step S6 and arrange inspections and maintenance as soon as possible to prevent the state from deteriorating further;

[0114] The alarm state indicates that the health condition of the device has seriously deteriorated, and the failure may occur immediately or has already occurred. The system must issue the highest-level alarm signal, such as triggering an audible and visual alarm, sending an emergency text message, etc., requiring the maintenance personnel to immediately stop the machine for inspection and carry out emergency repairs or fault handling to ensure the safety of the device and production.

[0115] S6. Based on the feature vector obtained by the decomposition of the principal eigenvalue, the root cause of the fault is located to obtain the main evaluation dimensions that lead to the decline of the comprehensive health index.

[0116] It is important to know that the core of this step S6 is to use the mathematical properties of principal component analysis to accurately locate the main fault root cause dimension that leads to the decline of the comprehensive health index from a data-driven perspective, thereby realizing the tracing from the overall status assessment to the specific problem dimension, and providing decision support for subsequent targeted maintenance.

[0117] Furthermore, in the above technical solution, the root cause localization of the fault is achieved by analyzing the dimension corresponding to the component with the largest absolute value in the feature vector corresponding to the principal feature value. Specifically, the operation is as follows:

[0118] S61. Extract the feature vector v1 corresponding to the principal feature value λ1;

[0119] S62. Calculate the absolute value of each component in the feature vector v1;

[0120] S63. Find the component with the largest absolute value;

[0121] S64. Determine the index of the component in the feature vector. This index is mapped back to the original evaluation dimension order. For example, index 1 corresponds to the electrical dimension, index 2 corresponds to the mechanical dimension, and so on, thereby identifying the main evaluation dimensions that lead to the deterioration of health status.

[0122] It's important to understand that this step, based on multivariate statistical principles, effectively isolates the coupling relationships between parameters across multiple dimensions, avoiding misjudgments caused by parameter correlations and thus more accurately pinpointing the core problem dimension. For example, if the absolute value of the component corresponding to the mechanical dimension in feature vector v1 is the largest, it indicates that the current degradation of the equipment's health is mainly due to abnormalities in mechanical components, such as bearing wear or gearbox failure, thus guiding maintenance personnel to prioritize checking the mechanical system.

[0123] S7. Based on the comparison results between the main evaluation dimensions and the maintenance feedback data, adaptively optimize the weight diagonal matrix and threshold.

[0124] Furthermore, in the above technical solution, the specific process of adaptive optimization is as follows:

[0125] S71. Record the causes of failures (R) after industrial equipment maintenance, the monitoring index data before and after maintenance, and the changes in the comprehensive health index (HI).

[0126] It's important to understand that this step aims to establish a maintenance case library for model optimization. The recording process is automated. When the monitoring system detects a maintenance event, such as a work order closing or the equipment status changing from "stopped" to "running," it automatically triggers the data archiving process. The fault cause R is selected by maintenance personnel from predefined options in the work order or manually entered. This fault cause is mapped to its corresponding evaluation dimension; for example, "bearing wear" is mapped to the mechanical dimension, and "drive failure" is mapped to the electrical dimension. The monitoring index data before and after maintenance refers to the health status matrix X collected at the time of the fault or the last time before maintenance. before The health status matrix X collected during stable operation of the equipment after maintenance after The change in the comprehensive health index HI is recorded. before and HI after All of this data will form a complete repair case and be stored in a historical database.

[0127] S72. Compare the main evaluation dimensions with the evaluation dimensions corresponding to the fault cause R, and count the number of positioning deviations.

[0128] It is important to understand that the consistency comparison specifically involves comparing the main evaluation dimensions identified in step S6 with the actual evaluation dimensions corresponding to the fault cause R in the maintenance record. If they match, it is counted as a correct positioning; if they do not match, it is counted as a positioning deviation. The system maintains a positioning deviation counter for each dimension, for example, Count. elec , Count mech , Count proc , Count env Each time a positioning deviation occurs, the counter for the corresponding dimension is incremented by 1. The count of positioning deviations refers to reading the current value of these counters.

[0129] S73. When the number of positioning deviations exceeds the preset threshold, the diagonal elements of the weight diagonal matrix are adjusted using the gradient descent algorithm.

[0130] It is important to know that the preset threshold is an empirical value, for example, it can be set to 5 times. When the number of positioning deviations in a certain dimension exceeds this threshold, it indicates that the weight currently assigned to that dimension may be unreasonable, leading to frequent misjudgments by the system, thereby triggering the optimization process. The core idea of ​​adjusting the diagonal elements of the weight diagonal matrix using the gradient descent algorithm is to construct a loss function to punish those cases in which the "fault root cause positioning dimension" and the "true fault dimension" are inconsistent in historical fault cases, and to iteratively optimize the weight diagonal matrix by minimizing this loss function. The loss function is constructed as follows: For each case k in the historical database where a positioning deviation occurs, obtain the true fault dimension of the case, denoted as jtrue(k). For example, when k=1, jtrue(1) is the electrical dimension. Then obtain the main evaluation dimension located in step S6, denoted as jpred(k). When jtrue(k) equals jpred(k), the loss L k A value of 0 indicates correct positioning with no loss; when jtrue(k) is not equal to jpred(k), there is a loss. The meaning of this loss term is: the weight w of the dimension of the actual fault. jtrue(k) The weight w of the main evaluation dimension identified in step S6 is less than the weight w of the main evaluation dimension identified in step S6. jpred(k) This will result in a large loss value, and the optimization process will increase w. jtrue(k) and / or reduce w jpred(k) To minimize this loss, thereby encouraging a greater tendency to use j in similar situations in the future. true(k) Dimension identification is the principal component. Finally, the losses of all historical bias cases are summed to obtain the total loss L = ΣL. k .

[0131] After adjusting the diagonal elements of the weight diagonal matrix using the gradient descent algorithm, the corresponding positioning deviation counter is reset to zero and starts counting again.

[0132] S74. Based on the normal fluctuation range of HI under different operating cycles or different environments of industrial equipment, dynamically update the threshold in step S5.

[0133] It's important to understand that the different operating cycles refer to equipment overhaul cycles, wear cycles, etc.; the different environments refer to different seasons, production of different products, etc. The normal baseline HI value of equipment will differ in the early, middle, and late stages of its lifespan. For example, the HI value of new equipment may remain stable above 95 for a long period, while it is normal for the HI value of older equipment to fluctuate between 80 and 85 when there are no faults. This step uses statistical analysis of historical data to dynamically update the threshold. The system periodically (e.g., quarterly) retrieves the HI values ​​of all health states over a past period (e.g., six months) and calculates their distribution, such as the mean μ and standard deviation σ. Then, based on the distribution, the threshold in step S5 is reset, for example:

[0134] The threshold a can be updated to μ-k a ×σ;

[0135] The threshold b can be updated to a-Δ1;

[0136] The threshold c can be updated to b-Δ2;

[0137] Where k a The multiplier is set based on experience, and its value ranges from 1 to 2. Δ1 and Δ2 are preset fixed interval values.

[0138] A health status assessment system for packaging machinery based on multi-source data fusion, such as Figure 2 As shown, it includes the following modules:

[0139] The data acquisition and feature construction module is used to collect operational data of packaging machinery in multiple preset dimensions, obtain data of at least two monitoring indicators for each dimension, and construct feature vectors corresponding to each dimension.

[0140] The data normalization module is used to combine the feature vectors corresponding to each dimension into a health status matrix and perform normalization processing to obtain a normalized matrix.

[0141] The weight allocation and weighted calculation module is used to allocate weights to each dimension, construct a weight diagonal matrix, and multiply the normalized matrix with the weight diagonal matrix to obtain a weighted state matrix;

[0142] The comprehensive health index calculation module is used to calculate the covariance matrix of the weighted state matrix, perform eigenvalue decomposition to obtain the principal eigenvalues, and then calculate the comprehensive health index based on the principal eigenvalues.

[0143] The health status assessment and early warning module is used to determine the health status of the equipment based on the comprehensive health index and to issue an early warning when the comprehensive health index is lower than a set threshold.

[0144] The fault root cause localization module is used to locate the fault root cause based on the feature vector obtained by the decomposition of the main feature value, and to obtain the main evaluation dimensions that lead to the decline of the comprehensive health index.

[0145] The adaptive optimization module is used to adaptively optimize the weight diagonal matrix and threshold based on the comparison results between the main evaluation dimensions and maintenance feedback data.

[0146] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A packaging machine health state evaluation method based on multi-source data fusion, characterized in that, The method comprises the following steps: S1, collecting running data of the packaging machine in multiple preset dimensions to obtain at least two monitoring index data in each dimension and to construct a feature vector corresponding to each dimension; The multiple preset dimensions are four evaluation dimensions, including electrical dimension, mechanical dimension, process dimension and environmental dimension; The monitoring index data of the electrical dimension includes motor current harmonic distortion rate THD and drive power factor PF, and the corresponding feature vector V elec =[THD, PF]; The monitoring index data of the mechanical dimension comprises an effective value V of driving end bearing vibration speed rms and a kurtosis value K of gearbox vibration signal urt , the corresponding feature vector V mech =[V rms , K urt ] The monitoring index data of the process dimension includes a unit-time qualified product rate P rat and a sealing temperature fluctuation standard deviation T std , a corresponding feature vector V proc = [P rat , T std ] The monitoring index data of the environment dimension includes main shaft bearing temperature T bea and environmental dust concentration D ust , corresponding feature vector V env = [T bea , D ust ] ; S2, combining the feature vector corresponding to each dimension into a health state matrix and performing normalization processing to obtain a normalized matrix; The normalization processing is realized by using a membership function to map elements of the health state matrix to the interval [0, 1]; wherein, the monitoring index data representing the better equipment state with the lower numerical value adopts a descending half trapezoidal membership function, and the monitoring index data representing the better equipment state with the higher numerical value adopts an ascending half trapezoidal membership function; S3, assigning weights to each dimension, constructing a weight diagonal matrix, and multiplying the normalized matrix by the weight diagonal matrix to obtain a weighted state matrix; S4, calculating the covariance matrix of the weighted state matrix, performing eigenvalue decomposition to obtain a principal eigenvalue, and then calculating a comprehensive health degree index according to the principal eigenvalue; The calculation formula of the comprehensive health degree index is: HI = 100 x (1 - λ1 / λ ideal ), Wherein, HI is the comprehensive health index, λ1 is the current main eigenvalue, λ ideal is the maximum main eigenvalue of the device in the brand new ideal state. S5, determining the equipment health state according to the comprehensive health degree index, and issuing a warning when the comprehensive health degree index is lower than a set threshold; S6, performing fault root cause positioning according to the feature vector obtained by the principal eigenvalue decomposition to obtain the main evaluation dimension causing the decrease of the comprehensive health degree index; S7, adaptively optimizing the weight diagonal matrix and the threshold according to the comparison result of the main evaluation dimension and the maintenance feedback data; The specific process of the adaptive optimization is: S71, recording the fault cause R of the industrial equipment after maintenance, the monitoring index data before and after maintenance, and the change of the comprehensive health degree index HI; S72, performing consistency comparison between the main evaluation dimension and the evaluation dimension corresponding to the fault cause R, and counting the positioning deviation times; S73, when the positioning deviation times exceed a preset threshold, adjusting the diagonal elements of the weight diagonal matrix by using a gradient descent algorithm; S74, dynamically updating the threshold in step S5 based on the normal fluctuation range of HI under different operation periods or different environments of the industrial equipment.

2. The packaging machine health state evaluation method based on multi-source data fusion according to claim 1, characterized in that, The determination of the equipment health state is specifically: When the comprehensive health degree index HI is greater than or equal to a, it is determined as a healthy state; When the comprehensive health degree index b is less than or equal to HI and less than a, it is determined as a sub-healthy state; When the comprehensive health degree index c is less than or equal to HI and less than b, it is determined as a warning state; When the comprehensive health degree index HI is less than c, it is determined as an alarm state.

3. The packaging machine health state evaluation method based on multi-source data fusion according to claim 1, characterized in that, The fault root cause positioning is realized by analyzing the dimension corresponding to the component with the largest absolute value in the feature vector corresponding to the principal eigenvalue.

4. The packaging machine health state evaluation system based on multi-source data fusion according to any one of claims 1-3, characterized in that, The method comprises the following modules: A data acquisition and feature construction module for collecting running data of the packaging machine in multiple preset dimensions to obtain at least two monitoring index data in each dimension and to construct a feature vector corresponding to each dimension; A data normalization processing module for combining the feature vector corresponding to each dimension into a health state matrix and performing normalization processing to obtain a normalized matrix; The weight distribution and weighted calculation module is configured to assign weights to each dimension, construct a weight diagonal matrix, and multiply the normalized matrix by the weight diagonal matrix to obtain a weighted state matrix; The comprehensive health index calculation module is configured to calculate a covariance matrix of the weighted state matrix, perform eigenvalue decomposition to obtain a main eigenvalue, and calculate a comprehensive health index according to the main eigenvalue; The health state evaluation and early warning module is configured to determine the equipment health state according to the comprehensive health index, and issue a warning when the comprehensive health index is lower than a set threshold; The fault root cause positioning module is configured to perform fault root cause positioning according to the eigenvector obtained by the main eigenvalue decomposition to obtain the main evaluation dimension that causes the comprehensive health index to decrease; The adaptive optimization module is configured to perform adaptive optimization on the weight diagonal matrix and the threshold according to a comparison result of the main evaluation dimension and the maintenance feedback data.

Images