Automatic optimization algorithm for workstations on the product assembly line

Abstract

This invention discloses an automatic workstation optimization algorithm for product assembly lines, relating to the field of product assembly line management technology. The algorithm includes the following steps: S1. Acquiring image / video data containing each work step and obtaining the work step time through work step identification; S2. Periodically calculating the average time and standard deviation of time fluctuation for each work step; S3. Dynamic workstation optimization; S4. Calculating the change in production line balance rate before and after dynamic adjustment. This invention solves the technical problem of "automatic workstation scheduling optimization on product assembly lines online." When applied to product assembly lines, this algorithm enables online automatic workstation scheduling optimization. By eliminating bottlenecks, it achieves balanced, flexible, and intelligent production processes, thereby improving production efficiency and quality, and ultimately driving cost reduction and efficiency improvement for enterprises.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of product assembly line management, and particularly relates to a work station automatic optimization algorithm for a product assembly line. BACKGROUND

[0002] During product assembly on an assembly line, once a bottleneck work station appears, the overall beat control of the assembly line is affected, and then the production (assembly) efficiency and quality are affected.

[0003] In the prior art, for the above situation, the work station scheduling is generally performed online by the management personnel on the assembly site, and then the bottleneck work station is cancelled to restore the normal beat of the assembly line. The work station optimization method for the assembly line is not intelligent enough, the analysis of real-time production data is not rigorous enough and lacks pertinence, the data analysis accuracy is insufficient, and the production process control is not accurate enough.

[0004] Therefore, the present application is proposed. SUMMARY

[0005] The present application aims to solve the above problems in the prior art and provides a work station automatic optimization algorithm for a product assembly line, which can improve production efficiency and quality through online automatic work station scheduling optimization.

[0006] To achieve the above purpose, the present application provides a work station automatic optimization algorithm for a product assembly line, which comprises the following steps:

[0007] S1. Collecting image / video data containing each work step and obtaining work step time through work step identification;

[0008] S2. Periodically calculating the average time consumption and time consumption fluctuation standard deviation of each work step, specifically:

[0009] S2.1. Work step Average time consumption formula:

[0010]

[0011] Wherein, : Average time consumption of the work step in the statistical period; : Total number of workers; : Number of workers performing the work step in the statistical period; : Average time consumption of the worker performing the work step; : Average time consumption of the worker performing the work step;

[0012] Average time consumption of the worker performing the work step; Average time consumption of the worker performing the work step; Average time consumption formula of the worker performing the work step:​​

[0013]

[0014] in, : The time taken to execute a single step; :Worker implement Number of steps; assuming the worker No execution record, then =0;

[0015] S2.2. Formula for the standard deviation of time fluctuation:

[0016]

[0017] in, The standard deviation of time consumption fluctuation reflects the degree of data dispersion. , , The definition is the same as above;

[0018] S3. Dynamic workstation optimization, specifically:

[0019] S3.1. Timeline Construction

[0020] The total timeline is generated by summing the time consumed in each step sequentially, using the following formula:

[0021] ( For the first (Time consumed per step)

[0022] S3.2. Calculation of division points

[0023] Based on the number of workers Divide the total timeline into equal parts The coordinates of the dividing points of the segment are:

[0024] ( )

[0025] S3.3. Process Node Matching

[0026] Traverse the equally divided points and select the nearest work step's end time as the work process node. The matching rule is as follows:

[0027]

[0028] S3.4. Compatibility Coefficient Assessment

[0029] Calculate the matching coefficient for each process node. ,in ;

[0030] when At that time, adjacent process segments are merged to generate special processes. By increasing the number of workstations, we ensured that the time required for this process was reduced to below Tavg.

[0031] Furthermore, an automatic workstation optimization algorithm for a product assembly line preferably includes:

[0032] S4. Calculation of the change in production line balance rate before and after dynamic adjustment, using the following formula:

[0033] .

[0034] Furthermore, the aforementioned automatic workstation optimization algorithm for a product assembly line further preferably includes, in step S2, dynamically marking abnormal work steps whose time exceeds a threshold by comparing the current time consumption with the historical average, as shown in the formula:

[0035]

[0036] in, : The average time spent within the current statistical period; Historical long-term average processing time; Threshold coefficient.

[0037] Compared with existing technologies, the algorithm of this invention, when applied to product assembly lines, enables online automatic workstation scheduling optimization. By eliminating bottleneck workstations, it achieves a balanced, flexible, and intelligent production process, thereby improving production efficiency and quality, and ultimately helping enterprises reduce costs and increase efficiency. Specifically:

[0038] 1. Improve overall production efficiency

[0039] Eliminating inconsistent production rhythms: By balancing the workload of each workstation, the speed of the entire production line is prevented from being slowed down by some workstations due to task overload (bottleneck), thus making the production rhythm more consistent.

[0040] Increased production capacity: After eliminating the bottleneck, the overall output rate of the production line changes from being determined by the slowest workstation to a higher level after coordination among all workstations, resulting in a significant increase in output per unit time.

[0041] 2. Optimize resource utilization

[0042] Reduce waiting time and waste: Avoid downstream workstations being idle due to upstream bottlenecks, and reduce waiting time for equipment or personnel.

[0043] Balancing manpower and equipment load: By redistributing tasks, avoid excessive fatigue or idle resources at certain workstations, thereby reducing operating costs.

[0044] 3. Reduce production costs

[0045] Shorten production cycles: Reduce the accumulation and stagnation of semi-finished products caused by bottlenecks, and speed up product delivery.

[0046] Reduce inventory backlog: After eliminating bottlenecks, the amount of work-in-process inventory decreases, reducing warehousing and capital occupation costs.

[0047] 4. Improve product quality

[0048] Reduce human error: Balancing workstation pressure can reduce the error rate caused by workers rushing to complete tasks.

[0049] Standardized processes: Optimized workstation tasks are more ergonomic and reduce the impact of fatigue on quality.

[0050] 5. Data-driven continuous improvement

[0051] Real-time monitoring and feedback: The algorithm combines the Internet of Things (IoT) to collect data in real time, continuously identify potential bottlenecks and dynamically optimize them.

[0052] Predictive maintenance: By analyzing data from workstation equipment, potential failure risks can be identified in advance, reducing unexpected downtime.

[0053] 6. Improve employee satisfaction

[0054] Reasonable allocation of workload: Avoid overworking some workstations for a long time, and reduce employee fatigue and turnover rate.

[0055] Clearly defined responsibilities: The optimized task allocation is clearer, reducing collaboration problems caused by chaotic processes. Detailed Implementation

[0056] The technical solution of the present invention will now be clearly and completely described with reference to the embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0057] As one embodiment of the present invention, the automatic optimization algorithm for workstations in a product assembly line provided by the present invention includes the following specific steps in its implementation:

[0058] Step 1: Image Acquisition and Preprocessing

[0059] Hardware deployment:

[0060] An industrial-grade Basler ace 2 camera was installed directly above each workstation on the assembly line, equipped with a ring LED fill light to eliminate shadow interference.

[0061] The camera is connected to the edge computing device NVIDIA Jetson AGX Xavier via Gigabit Ethernet to ensure a video stream transmission rate of ≥60 frames per second.

[0062] Image preprocessing:

[0063] Grayscale conversion and noise reduction: The acquired RGB image is converted to a grayscale image, and Gaussian filtering (kernel size 5×5) is applied to eliminate noise;

[0064] ROI extraction: Delineate the worker's operating area (Region of Interest, ROI) to reduce background interference;

[0065] Timestamp: Add a millisecond-level timestamp to each frame of the image, with an accuracy error of ≤10ms.

[0066] Step Two: Work Step Identification and Time Recording

[0067] Action recognition model training:

[0068] Data annotation: Collect video data containing different work steps (such as B1-part picking, B2-assembly operation), and use the LabelImg annotation tool to mark the bounding boxes of worker's hands, tools and parts;

[0069] Model selection: YOLOv8s model was adopted, and the input resolution was adjusted to 640×640 to balance accuracy and speed;

[0070] Training configuration: Transfer learning is performed on the COCO pre-trained model, with an initial learning rate of 0.01, a batch size of 16, and a training cycle of 100 rounds.

[0071] Real-time process detection:

[0072] The preprocessed image is input into the trained YOLOv8 model, and the output detection results include:

[0073] Tool categories (such as electric screwdrivers, clamps);

[0074] Part type (e.g., L1-bolt, L2-bearing);

[0075] Action status (e.g., "grabbing" or "assembly complete").

[0076] Step switching determination: When a change in the combination of a specific tool and part is detected (e.g., from the disappearance of the electric screwdriver to the appearance of the fixture), it is determined as a step switching.

[0077] Timestamp Records:

[0078] When a process step begins, the current system time is recorded as start_time;

[0079] When the end of a work step is detected, record end_time and calculate the duration: duration = end_time - start_time;

[0080] Data format example:

[0081] {

[0082] "worker_id": "G1",

[0083] "step_id": "B2",

[0084] "part_id": "L3",

[0085] "start_time": "2023-10-05 14:23:45.678",

[0086] "end_time": "2023-10-05 14:23:48.912",

[0087] "duration": 3.234

[0088] }

[0089] Step 3: Data Storage and Dynamic Analysis

[0090] The data storage and dynamic analysis module of this invention achieves persistent storage and intelligent analysis of assembly line data through a structured database and a real-time computing mechanism. Specific implementation methods are as follows:

[0091] 1. Data storage structure design

[0092] The work step operation record table is used to store detailed time information for workers to complete each work step, and includes the following core fields:

[0093] Record identifier: A unique auto-incrementing integer, used as the primary key for data retrieval and association;

[0094] Worker ID: A string with a length limit of 20 characters, following the "G+number" encoding rule (such as G1, G2), used to identify specific operators;

[0095] Step number: A string with a length limit of 20 characters, following the "B+number" encoding rule (such as B1, B2), to identify the current operation step;

[0096] Part Number: A string with a length limit of 20 characters, following the "L+number" encoding rule (such as L1, L2), to identify the type of part being operated on;

[0097] Start and end times: Date and time type with millisecond precision, recording the start and end times of the work step respectively, in the format of "year-month-day hour:minute:second.millisecond";

[0098] Duration: Decimal value type, accurate to three decimal places, in seconds, automatically calculated from the difference between the end time and the start time.

[0099] The workstation optimization record table is used to store historical data of dynamic adjustment plans and contains the following core fields:

[0100] Optimization identifier: A unique auto-incrementing integer that serves as the primary key to identify each optimization operation;

[0101] Timestamp: A standard date and time type that records the time when the optimization plan was generated;

[0102] Original workstation and new workstation numbers: strings with a length limit of 20 characters, recording the worker allocation before and after the adjustment respectively;

[0103] Associated step number: A string with a length limit of 20 characters, identifying the range of steps involved in the optimization (e.g., B5-B7).

[0104] Efficiency Improvement Rate: A decimal value, accurate to two decimal places, representing the expected percentage improvement in efficiency.

[0105] 2. Data Processing Flow

[0106] (1) Real-time data writing

[0107] Through the transaction batch commit mechanism, the work step records generated by the visual recognition module are written to the database at fixed intervals (such as 5 seconds) to ensure data integrity and writing efficiency.

[0108] Before writing data, perform integrity checks: verify whether the duration is positive, whether the worker and work step numbers conform to the coding standards, and transfer illegal data to the exception log table for manual review.

[0109] (2) Dynamic aggregation analysis

[0110] Periodic statistics: Automatically executes aggregate queries every hour to calculate the average time and standard deviation of time fluctuation for each step, which is used to identify efficiency bottlenecks;

[0111] Work steps Average time formula:

[0112]

[0113] in, : The average time taken for each step within the statistical period; Total number of workers; : Executed within the statistical period The number of workers in each step of the process; :Worker implement Average time spent on each step;

[0114] Worker Execution steps The formula for calculating the average time consumption is:

[0115]

[0116] in, : Time taken for a single execution of a process step (unit: seconds); :Worker implement Number of steps; assuming the worker No execution record, then =0;

[0117] Formula for standard deviation of time fluctuation:

[0118]

[0119] in, The standard deviation of time consumption fluctuation reflects the degree of data dispersion. , , The definition is the same as above;

[0120] Anomaly detection: By comparing the current time consumption with the historical average, abnormal steps whose time consumption exceeds a threshold (e.g., 120% of the average) are dynamically marked, triggering real-time alerts. Formula:

[0121]

[0122] in, : The average time spent within the current statistical period; Historical long-term average processing time; Threshold coefficient (default) );

[0123] Optimization effect tracking: Link the optimization record table with the work step operation record table, calculate the actual efficiency improvement rate, verify the effectiveness of the algorithm and support model iteration.

[0124] (3) Data fault tolerance mechanism

[0125] Local caching: When the network is interrupted, the edge device temporarily stores operation records for up to 24 hours, which are automatically synchronized to the central database after the connection is restored;

[0126] Resume interrupted data transmission: By recording the identifier of the last successfully written data, it ensures that only incremental data is synchronized after an interruption, avoiding duplication or omission.

[0127] 3. Data security and access control

[0128] Access permission layering:

[0129] The data writing end only has database insert permissions to prevent accidental modification of historical records;

[0130] The data analysis client is open with read-only permissions, supporting aggregate queries but prohibiting modification of the original data;

[0131] Encryption during transmission: The database connection is encrypted using Transport Layer Security (TLS 1.3) to prevent data from being intercepted or tampered with during transmission;

[0132] Storage encryption: Sensitive fields such as the mapping relationship between worker number and production line location are encrypted using the Advanced Encryption Standard (AES-256).

[0133] Step 4: Dynamic workstation optimization

[0134] Optimize the algorithm implementation:

[0135] Timeline Construction: The total timeline is generated by sequentially adding up the time consumed in each step. The formula is as follows:

[0136] ( For the first (Time consumed per step)

[0137] Equal division point calculation: based on the number of workers Divide the total timeline into equal parts The coordinates of the dividing points of the segment are:

[0138] ( )

[0139] Process node matching: Traverse the equally divided points and select the process node with the closest process end time. The matching rule is as follows:

[0140]

[0141] Dynamic adjustment strategy:

[0142] Coefficient of conformity assessment: Calculate the coefficient of conformity for each process node. ,in ;

[0143] Exception handling: When When this happens, the following operations are triggered:

[0144] Merge adjacent process segments to generate special processes. ;

[0145] Allocate additional workstations (workers) and have multiple people collaborate to ensure that the time taken for this process is reduced to T. avg the following;

[0146] Command issuance: The new workstation allocation plan is sent to the production line PLC controller via the MQTT protocol. Example format:

[0147] {

[0148] "command": "reassign",

[0149] "workers": ["G3", "G4"],

[0150] "steps": ["B5-B7"],

[0151] "effective_time": "2023-10-05 15:00:00"

[0152] }

[0153] Step 5: Visualization and Anomaly Monitoring

[0154] Real-time Kanban building:

[0155] Connect to a MySQL data source using Grafana and create the following monitoring panel:

[0156] Workstation efficiency heatmap: Displays time distribution by worker ID and work step ID;

[0157] Bottleneck process warning: A red warning is triggered when the time taken for a certain process step exceeds 20% of the historical average for three consecutive times.

[0158] Optimization effect comparison: Displays the change in production line balance rate before and after dynamic adjustment. The calculation formula is:

[0159]

[0160] Exception handling process:

[0161] Local alarm: Trigger an audible and visual alarm on the edge computing device and push a notification to the workshop supervisor's mobile phone;

[0162] Automatic logging: Records abnormal events to the database, including time, workstation, extent of exceedance, and handling measures;

[0163] Root cause analysis: Based on historical data, a decision tree model is trained to automatically recommend potential improvement measures (such as tool replacement and skills training).

[0164] This invention is not limited to the preferred embodiments described above. Anyone can derive other products in various forms under the guidance of this invention. However, regardless of any changes in shape or structure, any technical solution that is the same as or similar to this application falls within the protection scope of this invention.

Claims

1. An algorithm for automatic optimization of stations of a product assembly line, characterized in that Steps comprising: S1. Collecting image / video data containing each work step, and obtaining work step time through work step identification; S2. Periodically calculating the average time consumption and time consumption fluctuation standard deviation of each work step, specifically: S2.

1. Step Average time consumption formula: ； in, : The average time taken for each step within the statistical period; Total number of workers; : Executed within the statistical period The number of workers in each step of the process; :Worker implement Average time spent on each step; Workers Execution steps Average time consumption calculation formula: ； wherein, : Time spent per execution of a work step; Time spent per execution of a work step; : Worker Execution Number of executions of a work step; assuming a worker No execution record, then = 0; S2.

2. Time consumption fluctuation standard deviation formula: ； wherein : standard deviation of time-consuming fluctuation, reflecting data dispersion degree; , , : as defined above; S3. Dynamic work station optimization, specifically: S3.

1. Time axis construction Cumulatively add the time consumption of each work step to generate a total time axis, the formula is: for the first time​ S3.

2. Equidivision point calculation According to the number of workers The total timeline is equally divided into segments, with the equally divided point coordinates being: ( ) S3.

3. Process node matching Traverse the equidivision points, select the nearest work step end time as the process node, and the matching rule is: ； S3.

4. Fitting coefficient evaluation calculating a conformance coefficient for each process node wherein ; When adjacent process segments are merged, a special process is created by adding a station to ensure that the process duration falls below Tavg.

2. The automatic optimization algorithm for the station of the product assembly line according to claim 1, characterized in that Further comprising: S4. Calculation of the change of line balancing rate before and after dynamic adjustment, the formula is: 。 3. The automatic optimization algorithm for a workstation of a product assembly line according to claim 1 or 2, characterized in that: In the step S2, further dynamically mark the abnormal work step whose time consumption exceeds the threshold value by comparing the current time consumption with the historical average, the formula is: wherein, : average time consumption in the current statistical period; : historical long-term average time consumption; : threshold coefficient.

Images