A method for automatically generating a cold rolling work roll maintenance schedule

Abstract

The present application relates to the technical field of intelligent maintenance of cold rolling production equipment in the metallurgical industry, and particularly relates to a method for automatically generating a cold rolling work roll maintenance plan, which comprises real-time data acquisition; data cleaning and feature extraction; intelligent wear degree evaluation: real-time wear degree prediction based on a wear index calculation model based on rolling tonnage and rolling force average; residual life prediction: residual life prediction based on time series data using an LSTM neural network; risk level determination: risk rating generated by combining surface defect depth and density, dividing into three risk levels: normal / early warning / emergency; maintenance rule matching: calling a preset maintenance strategy library; multi-objective optimization decision; generating a maintenance work order; dynamic feedback adjustment; completing a closed loop of plan generation; the present application realizes automation, precision and dynamic optimization of the maintenance plan, thereby achieving the purposes of prolonging the service life of the work roll, reducing maintenance costs, reducing quality accidents and improving production efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent maintenance technology for cold rolling production equipment in the metallurgical industry, and in particular to a method for automatically generating maintenance plans for cold rolling work rolls. Background Technology

[0002] In the cold rolling process, the work rolls, as key components that directly contact the strip steel, directly affect product quality and production efficiency. Currently, steel companies generally manage work rolls using a combination of periodic maintenance and reactive repair, relying primarily on the experience of operators to formulate maintenance plans. This method has significant drawbacks: First, fixed-cycle maintenance cannot accurately reflect the actual wear condition of the work rolls, easily leading to over-maintenance or under-maintenance; second, manual judgment cannot fully consider the influence of multiple factors such as rolling process parameters and equipment operating status, resulting in insufficient scientific rigor in decision-making; third, when sudden roll surface damage occurs, the response speed is slow, potentially causing batch quality incidents.

[0003] Existing technologies include work roll replacement reminder systems based on rolling tonnage, but these only consider a single factor and have limited predictive accuracy. Others use vibration monitoring to determine roll surface condition, but these are not linked to production plans and lack practicality. Furthermore, traditional methods lack comprehensive optimization of maintenance costs, production losses, and quality risks, making it difficult to maximize economic benefits. With the development of intelligent manufacturing technology, there is an urgent need for a systematic solution that can monitor work roll condition in real time, intelligently predict lifespan, and automatically generate optimal maintenance plans. Summary of the Invention

[0004] This invention provides a method for automatically generating maintenance plans for cold rolling work rolls, solving problems such as reliance on manual experience, low planning efficiency, high roll consumption costs due to untimely maintenance, and frequent unplanned downtime in existing cold rolling work roll maintenance management. It provides an automatic maintenance plan generation method for cold rolling work rolls based on multi-source data fusion and intelligent decision-making. By collecting rolling process parameters, equipment status, and roll surface quality data in real time, and combining intelligent wear assessment, remaining life prediction, and multi-objective optimization algorithms, the method achieves automated, precise, and dynamic optimization of maintenance plans, thereby extending the service life of work rolls, reducing maintenance costs, decreasing quality accidents, and improving production efficiency.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] A method for automatically generating maintenance plans for cold rolling mill work rolls includes the following steps:

[0007] S1. Real-time data acquisition: Real-time acquisition of rolling mileage, cumulative tonnage, rolling force curve, bearing temperature, vibration spectrum, roll shape laser scanning data, surface defect images and historical maintenance records of the work roll;

[0008] S2, Data cleaning and feature extraction;

[0009] S3. Intelligent Wear Assessment: A wear index calculation model based on the average rolling tonnage and rolling force predicts real-time wear.

[0010] S4. Remaining lifetime prediction: The remaining lifetime is predicted based on time series data using an LSTM neural network.

[0011] S5. Risk Level Determination: A risk rating is generated by combining the depth and density of surface defects, and the risk is divided into three levels: normal / warning / emergency.

[0012] S6. Maintenance rule matching: Call the preset maintenance strategy library;

[0013] S7. Multi-objective optimization decision-making: balancing cost, efficiency, and quality factors;

[0014] S8. Generate maintenance work order: Output the specific maintenance tasks and time;

[0015] S9. Dynamic Feedback Adjustment: Optimize the plan in real time based on updated data;

[0016] S10, Complete the plan to generate a closed loop.

[0017] Furthermore, the maintenance rule matching specifically includes: performing dynamic matching when calling the maintenance rule library; triggering a grinding task when the wear index exceeds a preset threshold A; triggering a replacement assessment when the remaining lifespan is lower than a critical value B; and forcibly generating a shutdown inspection command when the risk rating reaches level three.

[0018] Furthermore, threshold A and critical value B are dynamically configured based on the material of the work roll.

[0019] Furthermore, the multi-objective optimization decision adopts a linear weighted model: minimizing the objective function min(α·maintenance cost + β·downtime loss + γ·quality risk coefficient), and is constrained by the production planning window, spare parts inventory status, and equipment availability.

[0020] Furthermore, the generation of maintenance work orders includes: outputting a personalized task list, which includes roller number, maintenance type, execution time window, required working hours and associated equipment number, wherein the execution time window is automatically calculated by matching production gaps using an LSTM neural network.

[0021] Furthermore, the dynamic feedback adjustment specifically means: when the newly added rolling data causes the state assessment result to deviate from the original plan by more than 10%, a plan reconstruction is triggered; when a sudden surface spalling or vibration abnormality is detected, an emergency work order is inserted and the current plan is interrupted.

[0022] Furthermore, when generating a maintenance work order, the task instructions are simultaneously pushed to the maintenance personnel's mobile terminal, an equipment occupancy notification is sent to the production scheduling system, and the spare parts management system is triggered to start the material preparation process.

[0023] Furthermore, the LSTM neural network incorporates transfer learning during training, using historical data from the hot rolling mill production line to pre-train the model, which is then fine-tuned under cold rolling conditions before deployment.

[0024] Furthermore, the execution time window is associated with grinding machine resources when it is generated: when a precision grinding task times out and is not allocated, the cross-workshop grinding machine resource bidding and allocation algorithm is automatically started.

[0025] Furthermore, the dynamic feedback adjustment uses the actual effect data after each maintenance execution as a reinforcement learning signal, and retrains it with the real-time updated data to achieve online adaptive adjustment of the maintenance plan.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] 1) Significantly improve maintenance efficiency and accuracy: Through automated data collection and intelligent analysis, the automation and intelligence of maintenance plan formulation have been realized, which greatly improves maintenance efficiency, shortens the maintenance plan formulation time from 4-6 hours of traditional manual decision-making to within 10 minutes, and improves the wear assessment accuracy to over 95%, effectively avoiding misjudgment and missed detection problems caused by human experience;

[0028] 2) Significantly reduce production costs: The evaluation method based on multi-source data fusion significantly improves the accuracy of status judgment. As verified by actual production lines, this invention can extend the average service life of work rolls by 30%-40%, reduce roll consumption costs by more than 25%, and reduce capacity loss caused by unplanned downtime by about 15%, saving more than 2 million yuan in maintenance costs per production line per year.

[0029] 3) Intelligent dynamic optimization production guarantee: The unique multi-objective optimization algorithm and real-time feedback mechanism maximize economic benefits. The dynamic adjustment mechanism ensures that the maintenance plan matches the actual production needs in real time. It can automatically match the production rhythm while ensuring the surface quality of the strip steel, reducing the impact of maintenance operations on capacity to 1 / 3 of the traditional method. It is especially suitable for high-precision, fast-paced modern cold rolling production lines.

[0030] 4) By collecting rolling process parameters, equipment operating status, and roll surface quality data in real time, and combining them with multi-objective optimization algorithms, the automatic formulation and dynamic adjustment of work roll maintenance plans are realized. This effectively solves the problems of low efficiency, poor accuracy, and delayed response in traditional manual maintenance planning. It provides an intelligent solution for the stable operation and cost reduction of cold rolling production lines, effectively extends the service life of work rolls, reduces maintenance costs, and reduces unplanned downtime, thus providing a strong guarantee for the stable and efficient operation of cold rolling production lines. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the method of the present invention. Detailed Implementation

[0032] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0033] See Figure 1 This is a schematic diagram of the method of the present invention. The present invention provides a method for automatically generating maintenance plans for cold rolling work rolls. It constructs a complete intelligent maintenance decision-making method, which achieves accurate assessment of roll status by real-time collection of work roll operating data and combining it with intelligent algorithms. Based on multi-objective optimization, it automatically generates the optimal maintenance plan, significantly improving the intelligence level of cold rolling work roll maintenance management. First, it collects rolling process parameters (including rolling mileage, tonnage, rolling force, temperature, etc.), equipment operating status (vibration spectrum, bearing temperature, etc.), and roll surface quality data (laser scan data, surface defect images, etc.) of the work rolls in real time through a multi-source sensor network. Second, it uses a hybrid evaluation method that integrates physical models and machine learning algorithms to accurately assess the real-time wear degree, remaining life, and potential risks of the work rolls. Then, based on a preset maintenance rule base and optimization objectives (including maintenance costs, downtime losses, and quality risks), it automatically generates personalized maintenance plans through an intelligent decision-making model. Finally, the system also has a dynamic adjustment function, which can optimize and adjust the maintenance plan according to real-time monitoring data and changes in production plans. Specifically, it includes the following steps:

[0034] S1. Real-time data acquisition: Real-time acquisition of rolling mileage, cumulative tonnage, rolling force curve, bearing temperature, vibration spectrum, roll shape laser scanning data, surface defect images and historical maintenance records of the work roll;

[0035] Force sensors with a range of 0-20MN and an accuracy of ±0.5%FS are installed on the mill stand to collect rolling force signals in real time.

[0036] The cumulative rolling kilometers of the work rolls are recorded using a 0.1m resolution encoder.

[0037] An infrared thermal imager with a frame rate of 30Hz was used to monitor the temperature field distribution on the roller surface;

[0038] An accelerometer with a frequency response of 0.5-10kHz was installed to collect vibration signals;

[0039] After every 5 rolls of steel are rolled, an installed laser scanner with an accuracy of ±2μm is automatically triggered to perform full roll surface inspection.

[0040] S2. Data cleaning and feature extraction: Preprocessing the collected data, including outlier cleaning, multi-source spatiotemporal alignment, and feature engineering extraction;

[0041] 1) Apply a moving average filter to the original data (window width 100ms);

[0042] 2) Remove outlier data points that exceed the 3σ rule;

[0043] 3) Extract the 1 / 3 octave band spectral features (16 frequency bands) of the vibration signal;

[0044] 4) Calculate 12 geometric parameters of the roll profile, including convexity and wedge shape;

[0045] 5) Quantify the proportion of surface defect area using image processing algorithms.

[0046] S3. Intelligent Wear Assessment: A wear index calculation model based on the average rolling tonnage and rolling force predicts real-time wear.

[0047] Input characteristics: rolling tonnage (normalized value), average rolling force (MPa), surface roughness Ra (μm);

[0048] Calculate the comprehensive wear index: W = 0.55 × tonnage + 0.35 × rolling force + 0.1 × Ra;

[0049] Grading criteria: W < 0.4 is considered normal; 0.4 ≤ W < 0.7 is considered a warning; W ≥ 0.7 is considered an emergency.

[0050] S4. Remaining life prediction: The remaining life is predicted based on time series data using an LSTM neural network, and the available cycle is predicted based on the LSTM model. The LSTM neural network introduces transfer learning during training, and the model is pre-trained using historical data from the hot rolling mill production line and then deployed after fine-tuning under cold rolling conditions.

[0051] LSTM model input dimensions: 16 (including the rolling parameter time series of the past 8 hours);

[0052] Network structure: two 128-unit hidden layers, dropout rate 0.2;

[0053] Output: Remaining rolling kilometers (95% confidence interval);

[0054] Online update mechanism: The model is retrained every 50 new sets of data.

[0055] S5. Risk Level Determination: A risk rating is generated by combining the depth and density of surface defects, and the risk is divided into three levels: normal / warning / emergency.

[0056] Level 1 is normal: wear degree W < 0.4 and no surface defects;

[0057] Level 2 is a warning: 0.4 ≤ W < 0.7 or fewer than 3 point defects;

[0058] Level 3 is an emergency: W ≥ 0.7 or a strip-shaped defect appears.

[0059] S6. Maintenance Rule Matching: Call the preset maintenance strategy library; the maintenance rule matching specifically includes: performing dynamic matching when calling the maintenance rule library, triggering a grinding task when the wear index exceeds the preset threshold A, triggering a replacement assessment when the remaining life is lower than the critical value B, and forcibly generating a shutdown inspection command when the risk rating reaches level three; threshold A and critical value B are dynamically configured according to the work roll material: for example, the threshold A of a tungsten carbide work roll is 25% higher than that of a high-chromium steel work roll, and the critical value B is 15% lower.

[0060] The rule base contains 32 expert rules, for example:

[0061] If risk level = 3 and remaining lifespan < 80km, then replace immediately.

[0062] If 0.6≤W<0.7 AND production gap>4h THEN, schedule grinding.

[0063] S7. Multi-objective optimization decision: balancing cost, efficiency, and quality factors; the multi-objective optimization decision adopts a linear weighted model: the objective function is Min(0.4×cost + 0.3×downtime + 0.3×quality risk), and is constrained by production planning window, spare parts inventory status, and equipment availability.

[0064] S8. Generate Maintenance Work Order: Output specific maintenance tasks and times; The generation of maintenance work order includes: outputting a personalized task list, which includes roller number, maintenance type, execution time window, required working hours and associated equipment number, wherein the execution time window is automatically calculated by matching production gaps using an LSTM model; It also simultaneously pushes task instructions to the maintenance personnel's mobile terminal, sends equipment occupancy notification to the production scheduling system, and triggers the spare parts management system to start the material preparation process; The execution time window is associated with grinding machine resources when it is generated: When a precision grinding task times out and is not allocated, the cross-workshop grinding machine resource bidding allocation algorithm is automatically started;

[0065] The output fields include:

[0066] Roller number: Automatically associated with RFID tags;

[0067] Maintenance type: Grinding / Replacement / Inspection;

[0068] Time window: accurate to 15 minutes;

[0069] Required resources: Grinding machine serial number, personnel qualification requirements;

[0070] Acceptance criteria: Roughness Ra≤0.5μm.

[0071] S9. Dynamic Feedback Adjustment: The plan is optimized in real time based on the updated data. Specifically, the dynamic feedback adjustment is as follows: when the newly added rolling data causes the deviation between the status assessment result and the original plan to exceed 10%, the plan is reconstructed; when a sudden surface spalling or vibration abnormality is detected, an emergency work order is inserted and the current plan is interrupted. Furthermore, the dynamic feedback adjustment uses the actual effect data after each maintenance execution as a reinforcement learning signal and retrains it with the updated data in real time to achieve online adaptive adjustment of the maintenance plan.

[0072] S10, Complete the plan to generate a closed loop.

[0073] The following embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments. Unless otherwise specified, the methods used in the following embodiments are conventional methods.

[0074] Example

[0075] This invention provides a method for generating an automatic maintenance plan for cold rolling work rolls, comprising the following steps: real-time acquisition of rolling process parameters, equipment operating status, and surface quality data of the work rolls via a multi-source sensor network; after preprocessing and feature extraction of the acquired data, a hybrid evaluation method integrating physical models and machine learning algorithms is used to calculate the real-time wear degree of the work rolls, predict their remaining life, and classify risks; based on a preset maintenance rule base, an optimal maintenance plan is automatically generated using a multi-objective optimization algorithm, the plan including maintenance type, execution time window, and resource requirements; a dynamic feedback mechanism is established to adaptively adjust the maintenance plan according to real-time monitoring data and changes in production plans; finally, a structured maintenance work order is output and automatically pushed to the execution system; through data-driven and intelligent decision-making, the precision and automation of work roll maintenance are achieved.

[0076] An edge computing-based intelligent maintenance system was deployed on a five-stand cold rolling mill production line, equipped with a 16-channel vibration monitoring module (50kHz sampling rate) and a high-definition roll surface scanner (5μm resolution). Data is transmitted in real time to a cloud analysis platform via industrial Ethernet. A predictive model is constructed using a hybrid LSTM+random forest algorithm, with a wear threshold of 0.65 set as the grinding trigger point. Preventive maintenance is automatically executed during the low-production period from 23:00 to 01:00 daily. Maintenance work orders are presented in 3D visualization through a digital twin system and are deeply integrated with the MES system to achieve closed-loop management from condition monitoring to planned execution. After this implementation was applied in a steel plant, the accidental damage rate of work rolls decreased by 72%, and the maintenance cost per ton of steel decreased by 18 yuan.

Claims

1. A method of automatically generating a cold rolling work roll maintenance schedule, characterized by, Comprise the following steps: S1, real-time data acquisition: real-time acquisition of rolling kilometers, cumulative tonnage, rolling force curve, bearing temperature, vibration spectrum, roll shape laser scanning data, surface defect image and historical maintenance record of work roll; S2, data cleaning and feature extraction; S3, intelligent wear degree evaluation: based on the wear index calculation model of rolling tonnage and rolling force average value to predict real-time wear degree; Calculate the comprehensive wear index: W=0.55×tonnage+0.35×rolling force+0.1×Ra; S4, residual life prediction: using LSTM neural network to predict residual life according to time series data; S5, risk level determination: combined with the depth and density of surface defects to generate risk rating, divided into three levels of risk: normal / early warning / emergency; S6, maintenance rule matching: calling the preset maintenance strategy library; S7, multi-objective optimization decision: balancing cost, efficiency and quality factors; The multi-objective optimization decision adopts a linear weighted model: min(α·maintenance cost+β·shutdown loss+γ·quality risk coefficient), and is limited by production plan window, spare parts inventory status and equipment availability constraints; S8, generating maintenance work order: output specific maintenance tasks and time; S9, dynamic feedback adjustment: real-time optimization of the plan according to real-time updated data; S10, complete plan generation closed loop.

2. The method of automatically generating a cold rolling work roll maintenance schedule of claim 1, wherein, The maintenance rule matching specifically includes: when calling the maintenance rule library, dynamic matching is performed, grinding task is triggered when the wear index exceeds the preset threshold A, replacement evaluation is triggered when the residual life is lower than the critical value B, and shutdown inspection instruction is forced to generate when the risk rating reaches level three.

3. A method of automatically generating a cold rolling work roll maintenance schedule according to claim 2, wherein, Threshold A and critical value B are dynamically configured according to the material of work roll.

4. The method of automatically generating a cold rolling work roll maintenance schedule of claim 1, wherein, The generation of maintenance work order includes: outputting an individualized task list, which contains roll number, maintenance type, execution time window, required man-hours and associated equipment number, wherein the execution time window is automatically calculated by the LSTM neural network matching production gap.

5. The method of automatically generating a cold rolling work roll maintenance schedule of claim 1 wherein, The dynamic feedback adjustment specifically is: when the state evaluation result deviates from the original plan by more than 10% due to the addition of rolling data, plan reconstruction is triggered, and when sudden surface spalling or vibration anomaly is detected, emergency work order is inserted and the current plan is interrupted.

6. The method of automatically generating a cold rolling work roll maintenance schedule of claim 1, wherein, When the maintenance work order is generated, the task instructions are also pushed to the mobile terminal of the maintenance personnel, the equipment occupation notice is sent to the production scheduling system, and the spare parts management system is triggered to start the material preparation process.

7. The method of automatically generating a cold rolling work roll maintenance schedule of claim 1 wherein, The LSTM neural network introduces transfer learning during training, pre-trains the model using hot continuous rolling line historical data, and deploys it after fine-tuning under cold rolling conditions.

8. The method of automatically generating a cold rolling work roll maintenance schedule of claim 4, wherein, The execution time window is associated with the grinding machine resource when it is generated: when the precision grinding task is not assigned in time, the cross-plant grinding machine resource bidding allocation algorithm is automatically started.

9. The method of automatically generating a cold rolling work roll maintenance schedule of claim 1 wherein, The dynamic feedback adjustment is to use the actual effect data after each maintenance execution as a reinforcement learning signal, and retrain with the real-time updated data to realize online adaptive adjustment of the maintenance plan.

Images