Assembly Line Real-Time Efficiency Evaluation System Based on Multi-Sensor Data Fusion

Abstract

This invention discloses a real-time assembly line efficiency evaluation system based on multi-sensor data fusion, specifically in the field of efficiency evaluation. The system includes a production information acquisition unit, a data matrix construction unit, an efficiency calculation and modeling unit, an efficiency analysis unit, and a decision execution unit. The production information acquisition unit collects process time, equipment running time, and workpiece pass rate. The data matrix construction unit, after format conversion and removal of outliers using a local anomaly factor algorithm, combines the data matrix to form a data matrix. The efficiency calculation and modeling unit processes the data matrix, calculates the efficiency vector norm using Minkowski distance, and then smooths out abnormal fluctuations using median filtering to obtain the efficiency vector. The efficiency analysis unit constructs an efficiency interval based on the efficiency vector and generates analysis results using a dynamic sliding window. The decision execution unit generates and issues management instructions based on the analysis results. This system achieves real-time and accurate evaluation of assembly line efficiency, providing a scientific basis for production management.

Description

Technical Field

[0001] This invention relates to the field of efficiency evaluation technology, and more specifically, to a real-time efficiency evaluation system for assembly lines based on multi-sensor data fusion. Background Technology

[0002] In assembly line production, efficiency assessment is a crucial step in ensuring production stability and increasing capacity. Existing technologies largely rely on manual recording or data collection from single sensors. For example, they rely on manually calculating process completion times and equipment downtime, or using only a single type of sensor to obtain partial production parameters, and then assess efficiency based on simple statistical methods (such as average working hours calculation).

[0003] However, significant shortcomings exist in practical applications: First, the data collection dimension is singular, failing to comprehensively reflect the assembly line's operational status. For example, focusing solely on processing time while ignoring equipment operational stability and workpiece quality leads to biased efficiency assessment results. Second, the data processing methods are crude, lacking an effective mechanism for removing abnormal data. Abnormal data generated by sensor malfunctions or sudden interference can easily lead to assessment biases. Third, efficiency analysis is lagging and lacks depth. Traditional methods are mostly post-event statistics, unable to capture efficiency trends in real time and struggling to quantify the degree of efficiency fluctuations, resulting in delayed production management decisions. Fourth, decision generation lacks systematicity, issuing instructions based solely on a single efficiency indicator, failing to combine multi-dimensional information such as efficiency levels and trends to formulate targeted solutions, making it difficult to effectively address production bottlenecks and hindering the overall improvement of assembly line operational efficiency. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a real-time assembly line efficiency evaluation system based on multi-sensor data fusion, which addresses the shortcomings of the prior art, such as one-sided data acquisition, coarse processing, delayed analysis, and lack of targeted decision-making, through the following solutions.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a real-time efficiency evaluation system for assembly lines based on multi-sensor data fusion, comprising: a production information acquisition unit for acquiring raw operating data and transmitting the raw operating data to a data matrix construction unit; the production information acquisition unit includes a production cycle time acquisition module, an equipment operation acquisition module, and a quality inspection acquisition module; the raw operating data includes process processing time, equipment operating time, and workpiece pass rate data;

[0006] Data matrix construction unit: Based on the original running data, a data matrix is ​​formed through format conversion and outlier sample removal methods, and the data matrix is ​​transmitted to the efficiency calculation and modeling unit; the data matrix construction unit includes a data format subunit, an outlier removal subunit, and a matrix combination subunit.

[0007] Efficiency calculation and modeling unit: Based on the data matrix, an efficiency vector is obtained through matrix normalization and vector norm calculation methods, and the efficiency vector is output to the efficiency analysis unit; the efficiency calculation and modeling unit includes a normalization processing subunit, an efficiency vector generation subunit, and an abnormal fluctuation smoothing subunit;

[0008] Efficiency Analysis Unit: Constructs statistical intervals based on the efficiency vector, divides efficiency levels through interval determination, identifies efficiency change trends by combining a sliding window mechanism, and finally generates efficiency evaluation results and transmits the efficiency evaluation results to the decision execution unit;

[0009] Decision execution unit: Used to generate and issue management instructions including production scheduling, equipment maintenance and cycle time optimization based on the analysis results, and execute instructions based on a priority mechanism that prioritizes equipment maintenance over production scheduling, and production scheduling over cycle time optimization.

[0010] The technical effects and advantages of this invention are as follows:

[0011] 1. This invention uses proximity switch sensors, micro switch sensors, and two sets of through-beam photoelectric sensors to collect data on process time, equipment running time, and workpiece pass rate, covering the core dimensions of assembly line production. At the same time, through redundant data collection in the same process, precise installation of sensors on equipment start-up and shutdown components, and setting signal stabilization time thresholds, it effectively avoids the limitations and false triggering problems of single sensor data collection, ensuring the integrity and accuracy of the original operating data and providing reliable data support for subsequent efficiency evaluation.

[0012] 2. In the data matrix construction unit of this invention, the data format subunit converts different types of raw data into a standardized format to eliminate differences in units and formats; the anomaly removal subunit adopts the local anomaly factor algorithm, which accurately identifies and removes abnormal samples that deviate from the normal production mode by calculating the relative relationship between the local density of data points and the density of nearest neighbor points, avoiding interference data from affecting the evaluation results, so that the constructed data matrix can truly reflect the assembly line operation status and provide a high-quality data foundation for subsequent efficiency calculations;

[0013] 3. The efficiency calculation modeling unit of this invention first adopts the quantile normalization method to map the data in each column of the data matrix to the same quantile interval, eliminating the influence of differences in the value range of different indicators on the calculation; then, it calculates the efficiency vector norm based on the Minkowski distance combined with the weight vector to ensure the scientific nature of the efficiency calculation; finally, it uses the median filtering algorithm to replace abnormal fluctuation data with the median value within the window, effectively removing efficiency data anomalies caused by random interference, and obtaining the final efficiency vector that can accurately reflect the actual efficiency of the assembly line.

[0014] 4. The efficiency analysis unit of this invention first constructs three levels of efficiency intervals (high, medium, and low) based on the mean and standard deviation of the efficiency vector to quickly determine the preliminary efficiency level; then, it adopts a dynamically adjustable sliding window to determine the magnitude of change (rising, stable, or falling) by calculating the slope of the efficiency vector within the window, and to determine the degree of fluctuation (small, normal, or large) by calculating the variance, thereby realizing multi-dimensional and real-time analysis of efficiency status. Compared with traditional post-event statistics, it can capture the pattern of efficiency changes more timely and comprehensively, helping managers to accurately locate efficiency problems.

[0015] 5. The decision execution unit of this invention generates differentiated management instructions (such as generating equipment maintenance instructions when efficiency declines and generating production scheduling instructions when efficiency is stable) by using an adaptive decision model and combining the three-dimensional analysis results of efficiency level, change trend, and fluctuation degree. At the same time, based on the priority adjudication mechanism of "equipment maintenance > production scheduling > cycle time optimization", it resolves instruction conflict issues, ensures that key production issues are handled first, effectively improves the production efficiency and stability of the assembly line, reduces operation and maintenance costs, and achieves scientific and efficient production management. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0017] Figure 2 This is a schematic diagram of the overall system flow of the present invention;

[0018] Figure 3 This is a flowchart of the LOF anomaly detection algorithm of the present invention. Detailed Implementation

[0019] 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.

[0020] like Figures 1 to 3 The assembly line real-time efficiency evaluation system based on multi-sensor data fusion shown includes:

[0021] Production information acquisition unit: used to collect raw operating data and transmit the raw operating data to the data matrix construction unit; the production information acquisition unit includes a production cycle acquisition module, an equipment operation acquisition module, and a quality inspection acquisition module; the raw operating data includes process processing time, equipment running time, and workpiece pass rate data;

[0022] Data matrix construction unit: Based on the original running data, a data matrix is ​​formed through format conversion and outlier sample removal methods, and the data matrix is ​​transmitted to the efficiency calculation and modeling unit; the data matrix construction unit includes a data format subunit, an outlier removal subunit, and a matrix combination subunit.

[0023] Efficiency calculation and modeling unit: Based on the data matrix, an efficiency vector is obtained through matrix normalization and vector norm calculation methods, and the efficiency vector is output to the efficiency analysis unit; the efficiency calculation and modeling unit includes a normalization processing subunit, an efficiency vector generation subunit, and an abnormal fluctuation smoothing subunit;

[0024] Efficiency Analysis Unit: Constructs statistical intervals based on the efficiency vector, divides efficiency levels through interval determination, identifies efficiency change trends by combining a sliding window mechanism, and finally generates efficiency evaluation results and transmits the efficiency evaluation results to the decision execution unit;

[0025] Decision execution unit: Used to generate and issue management instructions including production scheduling, equipment maintenance and cycle time optimization based on the analysis results, and execute instructions based on a priority mechanism that prioritizes equipment maintenance over production scheduling, and production scheduling over cycle time optimization.

[0026] It should be specifically noted that the production information acquisition unit collects process time, equipment running time, and workpiece pass rate data through the production cycle acquisition module, equipment operation acquisition module, and quality inspection acquisition module, respectively.

[0027] It should be further explained that the production cycle acquisition module uses a proximity switch sensor to acquire the processing time of each process; the equipment operation acquisition module uses a micro switch sensor to acquire the equipment running time; and the quality inspection acquisition module uses a dual photoelectric sensor + time / distance interval judgment method to statistically analyze the workpiece pass rate data.

[0028] It should be further explained that the processing time of the process is collected by installing proximity switch sensors at each workstation of the assembly line according to the conveying path and size of the workpiece. When the workpiece approaches the sensor with the conveyor belt, the sensor senses the workpiece and generates a sensing signal. The data acquisition terminal connected to the sensor is equipped with a high-precision timing chip, which can accurately record the time interval between two adjacent signal triggers. This time interval is the processing time of the current process. To further improve the accuracy and reliability of the data, multiple proximity switch sensors can be set up at the same workstation for redundant acquisition. When the time data error collected by multiple sensors is within the preset allowable range, the average of these data is taken as the final process processing time data.

[0029] It should be further explained that the method for collecting the device running time involves precisely installing microswitch sensors at key start-up and shutdown components of the device, such as the power switch and motor start button. This ensures that the sensors are closely linked to the device's start-up and shutdown actions. When the device starts running, the microswitch closes, and the sensor immediately sends a device running status signal to the data acquisition terminal. Upon receiving the signal, the data acquisition terminal starts timing. When the device stops, the microswitch opens, and the data acquisition terminal records the downtime and organizes the running time data. To prevent data errors caused by accidental triggering of the microswitch, a signal stabilization time threshold can be set on the data acquisition terminal. Only when the switch state remains consistently stable beyond this threshold is a change in device status confirmed, and the corresponding time information is recorded.

[0030] It should be further explained that the data collection method for the number of workpieces and the pass rate is as follows: On the workpiece conveyor path of the assembly line, two sets of through-beam photoelectric sensors are rationally planned and installed according to the size of the workpieces and the conveyor speed. One set of sensors is used to detect the number of workpieces. When a workpiece passes through the detection light curtain formed by this set of sensors with the conveyor belt, the sensor detects that the light is blocked and generates a signal change. The data acquisition terminal records the number of workpieces that have passed accordingly. The other set of sensors is used to detect the size parameters of the workpieces. By precisely arranging multiple sets of through-beam photoelectric sensors on both sides of the conveyor path, multiple detection light curtains are formed. When a workpiece passes through, the size data of the workpiece is calculated according to the different blocking conditions of the light curtains. The data acquisition terminal transmits the detected workpiece number and size data to the quality inspection and control unit. This unit has preset standard parameters for the number and size of workpieces. By comparing the actual detection data with the preset standards, it determines whether the workpiece is qualified. Within one inspection cycle, the number of qualified workpieces and the total number of inspected workpieces are counted to calculate the pass rate of the workpieces.

[0031] It should be further explained that the processing time of the process, the equipment running time, and the number and pass rate of the workpieces are integrated into the raw operating data and transmitted to the matrix construction unit.

[0032] It should be specifically noted that the data matrix construction unit plays a crucial role in connecting the preceding and following steps. It transforms the raw operational data transmitted by the production information acquisition unit into a data matrix through format conversion and abnormal sample removal, providing a standardized data foundation for the subsequent efficiency calculation and modeling unit.

[0033] It should be further explained that the data matrix construction unit consists of a data format subunit, an anomaly removal subunit, and a matrix combination subunit.

[0034] It should be further explained that the data format subunit will uniformly convert the processing time of each operation into numerical data in seconds. For example, if the production cycle time acquisition module collects the processing time of the i-th operation as... (The unit may originally be minutes or other time units). After processing by the data formatting subunit, it is converted into standard numerical data in seconds. For device running time data, this subunit will convert it into structured data of continuous running time, such as the time of the j-th device power-on. The downtime is Calculate the continuous running time If the equipment is not shut down (running continuously), the end time of the current statistical period will be used as the starting point. Finally, a structured data format is formed; for workpiece pass rate data, the data format sub-unit will convert it into percentage numerical data, assuming that the number of qualified workpieces is within a specific statistical period. The total number of inspected workpieces is The calculated workpiece pass rate This percentage value will be used as the standardized data format.

[0035] It should be further explained that the anomaly removal subunit receives standardized data processed by the data formatting subunit, including process processing time data in seconds, structured equipment running time data containing startup status indicators and continuous running duration, and percentage-based workpiece pass rate data. The data is first standardized to map data with different dimensions and value ranges to the same scale, eliminating the influence of dimensional differences on the calculation results. For example, for the process processing time data column... Equipment continuous running time data column and workpiece pass rate data column Z-score normalization formula was used respectively. Processing is carried out, among which This represents the mean of the data column. Let x be the standard deviation and x be the original data. This is the standardized data.

[0036] It should be further explained that when using the local anomaly factor algorithm, the key parameter k value needs to be determined, which is the number of nearest neighbors used to calculate the local reachability density. Based on the scale and distribution characteristics of the historical production data of the assembly line, and taking into account both computational efficiency and detection accuracy, a suitable k value can be determined through experiments or empirical rules. For example, when the amount of data is small, a smaller k value (such as k=5) can be selected; when the amount of data is large and the distribution is complex, the k value should be appropriately increased (such as k=10 or k=15).

[0037] It should be further explained that, for each data point p in the standardized dataset, the Euclidean distance between it and all other data points o in the dataset is calculated. Data points p and o represent vectors composed of production data in different dimensions. Assume the data has n dimensions (such as process time, equipment runtime, workpiece pass rate, etc.). Let the data points... , Then the Euclidean distance between them is calculated as follows: , where n represents the number of dimensions of the data, that is, the number of different production indicators contained in the data; It is the value of data point p in the i-th dimension. For example, when i=1 in the three dimensions including process time, equipment runtime, and workpiece pass rate, This represents the processing time of the process corresponding to data point p; It is the value of data point o in the i-th dimension. This formula measures the distance between two data points in a multidimensional space by calculating the square root of the sum of the squares of the differences between the data in each dimension.

[0038] Next, for each data point p, determine its k-th distance. That is, the distance from data point p to its k-th nearest neighbor. Based on the k-th distance, the reachability distance between data point p and data point o is calculated. The reachable distance takes into account both the local density of data point o and the distance relationship between p and o. and The larger value in the range is used as the reachability distance; for data point p, its local reachability density... ,in Let represent the set of k nearest neighbors of data point p, that is, the set of the k data points closest to data point p. This indicates the number of data points in the set. Local reachability density is the sum of reachable distances from data point p to its k nearest neighbors. It reflects the density of data points around data point p; a higher density indicates that the data distribution around p conforms more closely to the normal production model. After calculating the local reachability density, the local anomaly factor is further calculated using the following formula: The denominator of 1 is usually omitted. It is the average of the ratios of the local reachability density of p's k nearest neighbors to the local reachability density of p itself. Specifically, in the calculation, the local reachability density of each nearest neighbor o is first calculated. and The ratio reflects the relative relationship between the data density of the region containing the nearest neighbor point o and the data density of the region containing p. These ratios are then summed for all nearest neighbors and divided by the number of nearest neighbors. To obtain the average ratio, when When the value is close to 1, it indicates that the local density of p is similar to that of its nearest neighbors, and it is within the normal data distribution region; when... When the value is much greater than 1, it indicates that the local density of p is significantly lower than that of its nearest neighbors, which is likely an abnormal data point that deviates from the normal pattern.

[0039] It should be further explained that an anomaly detection threshold was subsequently set. This threshold can be adjusted based on factors such as assembly line production stability requirements and historical data statistical characteristics. When the local anomaly factor of data point p... If the data point is identified as an anomaly, the anomaly removal sub-unit will remove it from the dataset; if If so, then retain that data point.

[0040] It should be further explained that the threshold The setup involves collecting nearly three months of normal production data from the assembly line (at least 1000 valid records), standardizing it according to data format sub-unit rules, and calculating the data for each data point. Value (k value is determined by the amount of data: k=5 for m<1000, k=10 for m≥1000); for Sort the values ​​in ascending order and take the value corresponding to the 95th percentile as τ, i.e., P(LOF≤τ)=95%.

[0041] It should be further explained that the data retained after processing by the anomaly removal subunit will be integrated by the matrix combination subunit. The matrix combination subunit will use the standardized process processing time data column as the first column of the matrix, the structured data column of equipment running time containing the power-on status identifier and continuous running time as the second column, and the workpiece pass rate data column as the third column, and so on, combining the columns in sequence. Assuming that there are a total of m records after processing, the final result will be an m-row, 3-column data matrix. Its form is as follows:

[0042] Where m represents the total number of records in the processed data, used to determine the number of rows in the matrix, that is, there are a total of m production data records participating in the matrix construction. The processing time for the process corresponding to the i-th record, in seconds; The continuous runtime of the device corresponding to the i-th record, in seconds; The workpiece pass rate corresponding to the i-th record is presented as a percentage; finally, this matrix is ​​transmitted to the efficiency calculation and modeling unit.

[0043] It should be specifically noted that the efficiency calculation modeling unit receives the data matrix output by the data matrix construction unit and obtains the efficiency vector through a series of processes. It consists of a normalization processing subunit, an efficiency vector generation subunit, and an abnormal fluctuation smoothing subunit.

[0044] It should be further explained that the normalization processing subunit receives an m-row, 3-column data matrix M output by the data matrix construction unit. During matrix normalization, a quantile normalization method is used. This method maps each column of data in the data matrix to the same quantile interval to eliminate the influence of differences in dimensions and value ranges between different data columns on subsequent calculations. Specifically, for each column of data in the data matrix M (corresponding to process time, equipment running time, and workpiece pass rate data respectively), its quantile is calculated. Assuming the data is to be mapped to... For any data point x in a column of data, its quantile position is found, and its value x is converted to the corresponding interval. For example, if the quantile of data x in the column is q, then the value x is calculated based on the mapping relationship between quantiles and target intervals. This ensures that each column of data is uniformly within the same numerical range, resulting in a normalized data matrix. .

[0045] It should be further noted that the efficiency vector generation subunit is based on the normalized data matrix. A unified calculation method based on Minkowski distance is adopted to calculate the efficiency vector norm. The Minkowski distance formula is as follows: Where a and b are two data vectors, c is the vector dimension, and r is a parameter; the normalized data matrix Each row is considered a data vector, and a certain row of data vectors is used as the basis for further analysis. For example These are the normalized process time, continuous equipment operating time, and workpiece pass rate, respectively, combined with a preset weight vector. ( These correspond to the weights of process time, equipment runtime, and workpiece pass rate in efficiency evaluation, respectively. The efficiency vector norm corresponding to this row of data is calculated using the Minkowski distance formula; the specific calculation process is as follows: the data vector... With weight vector After multiplying corresponding elements, the efficiency vector norm corresponding to that row of data is obtained by calculating it according to the Minkowski distance formula. For data matrix Perform this calculation on each row to obtain a sequence of efficiency vector norms containing m elements. .

[0046] It should be further explained that the entropy weight method is used to determine the weight vector for an m-row, 3-column normalized data matrix. Perform positive normalization; calculate the information entropy of the j-th column index. in, (avoid =0); calculate weights Finally obtained ,satisfy .

[0047] It should be further explained that the value of the parameter r of the Minkowski distance is determined based on the data distribution characteristics. The default value of r = 2 (Euclidean distance) is suitable for most assembly line data (assuming that the data in each dimension follows a normal distribution). If the data in a certain dimension fluctuates greatly (such as the continuous running time of the equipment), r = 1 (Manhattan distance) can be adjusted to reduce the impact of outliers.

[0048] It should be further noted that the abnormal fluctuation smoothing subunit uses a median filtering algorithm to smooth the efficiency vector norm sequence. The median filtering algorithm processes the efficiency vector norm sequence by setting a window size s (e.g., s=3 or s=5, usually an odd number); specifically, it iterates through the efficiency vector norm sequence for each element. ,by Centered on i, select s elements (s elements before and after i) to form a window (if i is at either end of the sequence and there are fewer than s elements in the window, select according to the actual situation). Sort the elements in the window by value and take the value of the middle element as the center. Filtered result In this way, abnormal fluctuations in the efficiency vector norm sequence are removed, resulting in the final efficiency vector. The efficiency vector is then transmitted to the efficiency analysis unit.

[0049] It should be specifically noted that the efficiency analysis unit first processes the efficiency vector. Perform statistical analysis and calculate the average value of the vector. and standard deviation , where the average value ,in The efficiency value is the smoothed-out value after abnormal fluctuations, reflecting the overall efficiency level of the assembly line; standard deviation This measures the dispersion of efficiency data; subsequently, a three-level efficiency interval is constructed based on the mean and standard deviation: High-efficiency interval: This indicates that the efficiency is above average and the stability is good; the intermediate efficiency range is: This indicates that efficiency is within the normal fluctuation range; the inefficient range: This indicates that the efficiency is below average. This is determined by analyzing each element in the efficiency vector. The landing point range is used to determine the initial efficiency level (high efficiency, medium efficiency, or low efficiency) at the corresponding time.

[0050] It should be further explained that a dynamically resizing sliding window mechanism is used to analyze efficiency change trends. The window size *g* is quantitatively adjusted based on the process interval: when the process interval is <10 seconds (short interval), *g*=5; when the process interval is between 10-30 seconds (medium interval), *g*=8; when the process interval is >30 seconds (long interval), *g*=10. When the window covers edge data of the sequence (such as the 1st or mth data), a 'first / last data padding' strategy is adopted: if there are insufficient elements on the left side of the window, the 1st data is used to pad it; if there are insufficient elements on the right side of the window, the mth data is used to pad it, ensuring that the window always contains *g* elements and guaranteeing the accuracy of edge data calculation. The specific method for calculating the change amplitude is to analyze the efficiency data within the window. Perform linear fitting to obtain the slope ,like This is judged as an upward trend in efficiency; if If it is determined to be a stable trend; The trend is determined to be downward; the specific method for calculating the degree of fluctuation is to calculate the variance of the efficiency data within the window. in, These are the smoothed efficiency vector elements within the window. (The mean of the data within the window), if If the fluctuation is small; If the fluctuation is considered normal; It was determined to be highly volatile.

[0051] It should be further explained that the efficiency level, the magnitude of change and the degree of fluctuation are integrated in three dimensions to form a structured analysis result. High efficiency - rising - small fluctuation indicates that the current efficiency is excellent and continues to improve; medium efficiency - stable - large fluctuation indicates that the efficiency is normal but the stability is insufficient; low efficiency - declining - normal fluctuation indicates that the efficiency is low and is showing a deteriorating trend. The final analysis result is transmitted to the decision execution unit.

[0052] It should be noted that the decision-making execution unit first performs format standardization processing on the received analysis results, converting the three-dimensional description of "efficiency level - change range - fluctuation degree" into a quantitative feature vector. Where: L is the efficiency level quantification value (high efficiency = 3, medium efficiency = 2, low efficiency = 1); T is the change trend quantification value (rising = 1, stable = 0, falling = -1); V is the fluctuation degree quantification value (small fluctuation = 1, normal fluctuation = 0, large fluctuation = -1).

[0053] It needs further explanation that a hierarchical comprehensive judgment method is adopted for the quantified feature vector. Specifically, it includes the following: The first layer is based on the overall judgment of efficiency level L: when L=1, the system directly enters the anomaly handling process; when L=2, the system enters the optimization and maintenance balance process; when L=3, the system enters the stable operation guarantee process. The second layer further subdivides each overall judgment by combining the trend T and the degree of fluctuation V: when L=1 (inefficient), if T=-1 or V=-1, equipment maintenance instructions are generated; if T=0 and V=0, production scheduling and cycle time optimization instructions are generated; if T=1 and V=0, equipment maintenance instructions are generated. =1, generate a combined instruction (production scheduling + cycle time optimization + equipment maintenance); if L=2 (medium efficiency), if T=−1 or V=−1, generate an equipment maintenance instruction; if T=0 and V=0, generate production scheduling and cycle time optimization instructions; if T=1 or V=1, generate a combined instruction (production scheduling + cycle time optimization + equipment maintenance); when L=3 (high efficiency), if T=-1 or V=-1, generate an equipment maintenance instruction; if T=0 and V=0, generate production scheduling and cycle time optimization instructions; if T=1 and V=1, generate a combined instruction (production scheduling + cycle time optimization + equipment maintenance).

[0054] It should be further explained that when L=1 (inefficient): if T=−1 or V=−1, an equipment maintenance instruction is generated, which includes: fault risk diagnosis, key component maintenance plan, preventive maintenance arrangement, and abnormal operating condition alarm push; if T=0 and V=0, a production scheduling and cycle time optimization instruction is generated, which includes: adjusting the sequence of processes, balancing production load, and optimizing worker shifts; if T=1 and V=1, a joint instruction is generated, that is, the production scheduling and cycle time optimization instruction and the equipment maintenance instruction are issued simultaneously, which includes: implementing fine-tuning of production cycle time and optimizing capacity allocation on the premise of ensuring gradual equipment recovery.

[0055] It should be further explained that when L=2 (medium efficiency): if T=−1 or V=−1, an equipment maintenance instruction is generated, which includes: performance degradation monitoring, suggestions for replacing vulnerable parts, and calibration of equipment operating parameters; if T=0 and V=0, a production scheduling and cycle time optimization instruction is generated, which includes: maintaining the existing production cycle time, optimizing production scheduling in conjunction with market orders, and improving overall resource utilization; if T=1 or V=1, a joint instruction is generated, which includes: while maintaining cycle time optimization, strengthening equipment operation monitoring to avoid potential hidden dangers that lead to efficiency decline.

[0056] It should be further explained that when L=3 (high efficiency): if T=−1 or V=−1, an equipment maintenance instruction is generated, which includes: focusing on monitoring the wear of key parts, setting the maintenance time point and outputting early warning prompts; if T=0 and V=0, a production scheduling and cycle time optimization instruction is generated, which includes: maintaining high efficiency, appropriately compressing redundant time, and improving output per unit time; if T=1 or V=1, a joint instruction is generated, which includes: while monitoring and maintaining equipment status, continuously optimizing production scheduling and cycle time to ensure the stability of high-efficiency operation.

[0057] It should be further explained that the third layer: based on preset priorities, it makes decisions on multiple types of instructions: equipment maintenance instructions have the highest priority, followed by production scheduling instructions, and finally cycle time optimization instructions; if there are multiple conflicting instructions, the system will only issue the highest priority instruction according to the priority, and retain the other instructions in the form of a prompt message.

[0058] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0059] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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. An assembly line real-time efficiency evaluation system based on multi-sensor data fusion, characterized in that, The system comprises: a production information acquisition unit for acquiring original operation data and transmitting the original operation data to a data matrix construction unit; the production information acquisition unit comprises a production rhythm acquisition module, a device operation acquisition module, and a quality detection acquisition module; the original operation data comprises process processing time, device operation time, and workpiece qualification rate data; a data matrix construction unit for forming a data matrix based on the original operation data through format conversion and abnormal sample elimination, and transmitting the data matrix to an efficiency calculation modeling unit; the data matrix construction unit comprises a data format subunit, an abnormal elimination subunit, and a matrix combination subunit; the data format subunit is configured to convert the process processing time into numerical data in seconds, convert the device operation time data into structured data of continuous operation duration, and convert the workpiece qualification rate data into percentage numerical data; the abnormal elimination subunit is configured to identify and eliminate data samples deviating from the normal production mode based on a local anomaly factor algorithm; and the matrix combination subunit is configured to combine the processed original operation data to form a data matrix; an efficiency calculation modeling unit for obtaining an efficiency vector based on the data matrix through matrix normalization and vector norm calculation methods, and outputting the efficiency vector to an efficiency analysis unit; the efficiency calculation modeling unit comprises a normalization processing subunit, an efficiency vector generation subunit, and an abnormal fluctuation smoothing subunit; an efficiency analysis unit for constructing a statistical interval based on the efficiency vector, dividing efficiency levels through interval determination, and identifying efficiency change trends through a sliding window mechanism, and finally generating an analysis result and transmitting the analysis result to a decision execution unit; the analysis result is obtained by first calculating the average value and standard deviation of the efficiency vector after receiving the efficiency vector, and constructing high, medium, and low efficiency intervals, to determine the preliminary efficiency level by judging the landing point of the efficiency vector; then a sliding window with a dynamically adjusted size is used to determine the change amplitude by calculating the slope of the efficiency vector within the window, and to determine the fluctuation degree by calculating the variance of the efficiency vector within the window; finally, the efficiency level, change amplitude, and fluctuation degree are integrated into the analysis result; a decision execution unit for generating and issuing management instructions including production scheduling, device maintenance, and rhythm optimization based on the analysis result, and executing the instructions based on the priority mechanism that device maintenance is prior to production scheduling, and production scheduling is prior to rhythm optimization; the management instructions are obtained by first standardizing the analysis result to convert it into quantitative information containing efficiency level, change trend, and fluctuation degree; then based on the quantitative information, an adaptive decision model is used to generate management instructions suitable for the assembly line, covering production scheduling, device maintenance, and rhythm optimization; finally, when multiple generated instructions exist in execution conflicts, the highest priority instruction is executed preferentially according to the priority rules that device maintenance is prior to production scheduling, and production scheduling is prior to rhythm optimization, while the remaining instructions are retained.

2. The real-time efficiency evaluation system of assembly line based on multi-sensor data fusion according to claim 1, characterized in that: The production rhythm acquisition module adopts a proximity switch sensor to generate an induction signal when the workpiece approaches, and calculates the process processing time by recording the signal trigger interval; the equipment operation acquisition module uses a micro switch sensor installed on the equipment start-stop component, and the micro switch is closed when the equipment is running, and is opened when the equipment is stopped, and the equipment running time is obtained; the quality detection acquisition module includes two groups of reflection type photoelectric sensors arranged along the workpiece conveying path, which are used to detect the number and size parameters of the workpiece, and compare the detection results with the preset standard to judge and count the qualified rate of the workpiece.

3. The real-time efficiency evaluation system for assembly line based on multi-sensor data fusion of claim 1, wherein: The local anomaly factor algorithm determines the parameter k value, calculates the distance between data points, the k distance, the reachable distance, the local reachable density and the local anomaly factor, sets an anomaly judgment threshold, and identifies and removes data samples deviating from the normal production mode.

4. The real-time efficiency evaluation system for assembly line based on multi-sensor data fusion of claim 1, wherein: The function of the normalization processing subunit is to map each column of data in the data matrix to the same quantile interval during matrix normalization using the quantile normalization method; The function of the efficiency vector generation subunit is to calculate the efficiency vector norm using the weighted Minkowski norm calculation method; the function of the abnormal fluctuation smoothing subunit is to process the efficiency vector norm using the median filtering algorithm to remove abnormal fluctuation data and obtain the final efficiency vector; finally, the efficiency vector is transmitted to the efficiency analysis unit.

5. The multi-sensor data fusion based real-time efficiency evaluation system of an assembly line according to claim 4, wherein: The median filtering algorithm sets a sliding window for each position in the sequence in turn when processing the efficiency vector norm, arranges the data in the window in order and selects the central value to replace the original value at that position, and finally smooths the efficiency vector norm and removes abnormal fluctuation data.

Images