A method for integrated maintenance of a two-stage model server for rolling

Abstract

The application discloses a kind of integrative maintenance methods of two-stage model server for rolling steel, comprising S1, constructs layered heterogeneous hardware integration architecture: including virtualization and containerization deployment and hybrid cloud architecture design;S2, establish unified management and monitoring platform: including multidimensional monitoring system and automated operation mechanism;S3, implement process domain grouping and service management: including logical grouping and coordination mechanism and model service encapsulation.The application has the advantages that through hardware architecture optimization, management platform integration and operation process reengineering, the centralized management and efficient maintenance of server are realized, and the operation efficiency is improved, and production stability is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of digital operation and maintenance technology in the steel rolling industry, and particularly relates to an integrated maintenance method for a secondary model server used in steel rolling. Background Technology

[0002] In the steel rolling production process, the secondary model server undertakes core functions such as rolling force calculation, shape control, and process parameter optimization. However, existing technologies have the following problems: (1) Servers are deployed in various process links, forming "information islands". Maintenance personnel need to frequently switch between different management interfaces, resulting in low maintenance efficiency; (2) Uneven resource utilization, with some servers running at high load for a long time while other servers have idle resources, resulting in high hardware costs and energy consumption; (3) The lack of a unified monitoring and alarm mechanism makes it difficult to locate faults, leading to fluctuations in process parameters or even production shutdowns; (4) Model iteration and deployment rely on manual operation, version management is chaotic, and there is a delay bottleneck in cross-process data interaction. Summary of the Invention

[0003] The purpose of this invention is to solve the problems of existing servers being scattered and having low maintenance efficiency. It provides an integrated maintenance method for a secondary model server used in steel rolling. Through hardware architecture optimization, management platform integration, and operation and maintenance process reengineering, it achieves centralized management and efficient maintenance of the server, thereby improving operation and maintenance efficiency and ensuring production stability.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: An integration and maintenance method for a secondary model server used in steel rolling mills, specifically including: S1, building a layered heterogeneous hardware integration architecture: S11, Virtualization and Containerization Deployment: For non-core models with real-time requirements ≤100ms, physical server resources are pooled through a virtualization platform; for microservice models that support distributed deployment, container clusters are used for container encapsulation and cluster management. S12, Hybrid Cloud Architecture Design: The core model of the local hyperconverged architecture deployment has a real-time requirement of >100ms; non-real-time tasks are migrated to the enterprise private cloud or hybrid cloud, and cloud GPU resources are used to accelerate model iteration. S2, establish a unified management and monitoring platform: S21, a multi-dimensional monitoring system: collects server hardware indicators and model operating status, realizes visual monitoring through a unified monitoring module, and sets dynamic threshold alarms; at the same time, it builds a CMDB asset database to record the relationship between server-model-process, and supports fault tracing time ≤10min; S22, Automated Operation and Maintenance Mechanism: Using Terraform / Ansible to build Infrastructure as Code, it can complete the configuration, deployment and version rollback of 100+ servers with one click, reducing the deployment time from 4 hours to 20 minutes; In addition, it develops a model self-diagnosis tool, regularly injects boundary value test data, automatically verifies model output errors, and detects algorithm logic anomalies in advance. S3, implements process domain grouping and service-oriented management: S31, Logical Grouping and Collaboration Mechanism: Logical process groups are divided according to hot rolling, cold rolling, and heat treatment to collaboratively optimize process model parameters; at the same time, a gray-scale release strategy is adopted, with the new version model first running on 10%~20% of servers for 24 hours to verify that there are no abnormalities before being fully rolled out to reduce the risk of production interruption. S32, Model Service Encapsulation: The model is encapsulated as a RESTful interface through the API gateway for cross-process data calls, with communication latency controlled within 50ms; at the same time, version number rules are defined to implement version tracking of model code and parameter files based on Git, supporting fast rollback of any historical version within 15 minutes.

[0005] Furthermore, in step S11, the virtualization and containerization deployment specifically includes: (1) For the roll gap verification model, physical server resources are pooled through VMware vSphere or KVM virtualization platform, with a single physical machine carrying ≥8 virtual machines, improving resource utilization by 40%~60%; (2) The roll wear prediction model is encapsulated in Docker containers and managed by Kubernetes clusters to control cross-container communication latency at the 10ms level, supporting elastic scaling of 500+ containers.

[0006] Furthermore, in step S12, the hybrid cloud architecture design specifically includes: (1) Deploy the rolling process calculation model using Nutanix to ensure local data interaction latency <5ms; (2) Model training and historical data archiving are migrated to the enterprise private cloud or hybrid cloud, and the model iteration is accelerated by utilizing cloud GPU resources, which reduces training time by 30% to 50%.

[0007] Furthermore, in step S21, the server hardware indicators include CPU, memory, and disk I / O; the model running status includes computation time and convergence rate; the unified monitoring module is Prometheus+Grafana; the dynamic threshold alarm is based on the 3σ rule; and a level 1 alarm, i.e., model crash, will trigger a switch to a backup model within 500ms.

[0008] Furthermore, in step S22, Terraform is responsible for Infrastructure as Code (IaC), and Ansible is responsible for Configuration as Code (MCC). The rolling force calculation error is ≤±1%, and algorithm logic anomalies can be detected 72 hours in advance.

[0009] Furthermore, in step S31, each logic process group is assigned a responsible maintenance personnel, and a cross-group joint debugging meeting is established, with the frequency controlled at once per week.

[0010] Furthermore, in step S32, cross-process data retrieval includes calling hot-rolled roll wear data from the cold-rolled sheet shape model.

[0011] Furthermore, in steps S1 to S3, the virtualization platform VMware / KVM, the container cluster Docker / Kubernetes, the hyperconverged architecture Nutanix, and cloud resources constitute the hardware layer to realize the pooled management of server resources.

[0012] Furthermore, in steps S1 to S3, the unified monitoring module Prometheus+Grafana, the configuration management module CMDB+IaC, and the automated operation and maintenance module form the management layer, providing full lifecycle management capabilities.

[0013] Furthermore, in steps S1 to S3, the Model Service Bus (MSB) based on process domain grouping forms the application layer, which supports API calls and version management across process models and connects to the steel rolling production control system.

[0014] The system involved in the integration and maintenance method of the secondary model server for steel rolling of the present invention includes: Hardware layer: Composed of virtualization platforms such as VMware / KVM, container clusters such as Docker / Kubernetes, hyperconverged architecture such as Nutanix, and cloud resources, it enables pooled management of server resources; Management layer: Includes a unified monitoring module (Prometheus + Grafana), a configuration management module (CMDB + IaC), and an automated operation and maintenance module (script tools + fault self-healing), providing full lifecycle management capabilities; Application layer: Model Service Bus (MSB) based on process domain grouping, supports API calls and version management across process models, and interfaces with the steel rolling production control system.

[0015] Compared with the prior art, the advantages of the technical solution of the present invention are as follows: (1) This invention can effectively improve operation and maintenance efficiency. Through unified platform management, it can reduce cross-system switching operations by 70% and reduce daily maintenance time by 40% to 50%. (2) This invention optimizes resources, reduces hardware procurement costs by 50% to 70%, reduces server energy consumption by more than 30%, and increases resource utilization from 30% to more than 80%. (3) The method of the present invention enhances production stability, reduces fault recovery time from 4 hours to 1.5 hours, reduces model service interruption time by an average of 85% per year, and reduces process parameter fluctuation by 20%; (4) The method of the present invention has compatibility and scalability, supports a smooth transition between old and new systems, provides standardized interfaces for advanced applications such as digital twins and AI optimization, and reduces expansion costs by 60%. Detailed Implementation Example

[0016] To make the present invention clearer, the following description further illustrates the integration and maintenance method of a secondary model server for steel rolling. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the present invention.

[0017] This embodiment provides an integrated maintenance method for a secondary model server used in steel rolling, characterized in that: (a) Hardware integration implementation: (1) Equipment selection: Physical servers: Two Dell PowerEdge R750 servers, each equipped with four Intel Xeon Platinum 8480+ processors, 512GB DDR5 memory, and 8TB NVMe storage, supporting hardware virtualization acceleration VT-x / VT-d; Virtualization platform: KVM+QEMU, with OpenStack for resource scheduling, a single physical machine runs 8 virtual machines, each allocated 4 vCPUs / 32GB memory / 1TB storage; Network architecture: Configured with Huawei CE6850 switches, using SR-IOV technology to achieve 10Gbps pass-through between virtual machines, with network latency <50μs.

[0018] (2) Migration steps: The CentOS 7 system on 8 physical servers was migrated to KVM virtual machines without loss of data transfer using the virt-v2v tool, while retaining the original RAID configuration. Deploy Prometheus node exporter to collect server hardware metrics and build a resource utilization dashboard using Grafana; The ansible-playbook script is used to configure virtual machine network policies in batches to enable cross-VLAN communication.

[0019] (3) Validation data: Resource utilization: Before integration, the average load of a single server was 25%, and after integration it reached 75%, with CPU core utilization increasing from 12% to 60%. Energy consumption comparison: The annual power consumption of 2 physical servers is about 18,000 kWh, which is 73.5% less than the original 68,000 kWh of 8 servers, resulting in annual electricity savings of about 120,000 yuan.

[0020] (II) Deployment of the monitoring system (1) Indicator collection: Hardware layer: CPU temperature, fan speed, and power status are obtained via IPMI; Model layer: The Prometheus client is embedded in the rolling force calculation model to collect calculation time, number of iterations, and convergence status; Network layer: Use Prometheus' node_exporter to monitor switch port traffic and packet loss rate.

[0021] (2) Alarm strategy: Level 1 alert, i.e., model crash: triggers Zabbix to send an SMS to operations personnel, and simultaneously restarts the containerized model service via Kubernetes; Level 2 alarm, i.e., resource overload: When CPU utilization > 85% for 10 minutes, the virtual machine is automatically migrated to a standby physical machine in the resource pool. Level 3 alarms refer to data latency: Grafana threshold alarms automatically switch to local cached data when cross-process data interaction latency exceeds 50ms.

[0022] In addition to the embodiments described above, the present invention may have other implementations. All technical solutions formed by equivalent substitution or equivalent transformation fall within the protection scope claimed by the present invention.

Claims

1. An integration and maintenance method for a secondary model server used in steel rolling, characterized in that: S1, building a layered heterogeneous hardware integration architecture: S11, Virtualization and Containerization Deployment: For non-core models with real-time requirements ≤100ms, physical server resources are pooled through a virtualization platform; for microservice models that support distributed deployment, container clusters are used for container encapsulation and cluster management. S12, Hybrid Cloud Architecture Design: The core model of the local hyperconverged architecture deployment has a real-time requirement of >100ms; non-real-time tasks are migrated to the enterprise private cloud or hybrid cloud, and cloud GPU resources are used to accelerate model iteration. S2, establish a unified management and monitoring platform: S21, a multi-dimensional monitoring system: collects server hardware indicators and model operating status, realizes visual monitoring through a unified monitoring module, and sets dynamic threshold alarms; at the same time, it builds a CMDB asset database to record the relationship between server-model-process, and supports fault tracing time ≤10min; S22, Automated Operation and Maintenance Mechanism: Using Terraform / Ansible to build Infrastructure as Code, it can complete the configuration, deployment and version rollback of 100+ servers with one click, reducing the deployment time from 4 hours to 20 minutes; In addition, it develops a model self-diagnosis tool, regularly injects boundary value test data, automatically verifies model output errors, and detects algorithm logic anomalies in advance. S3, implements process domain grouping and service-oriented management: S31, Logical Grouping and Collaboration Mechanism: Logical process groups are divided according to hot rolling, cold rolling, and heat treatment, and process model parameters are collaboratively optimized; at the same time, a gray-scale release strategy is adopted, and the new version model is first piloted on 10%~20% of servers for 24 hours, and after verification that there are no abnormalities, it is fully promoted. S32, Model Service Encapsulation: The model is encapsulated as a RESTful interface through the API gateway for cross-process data calls, with communication latency controlled within 50ms; at the same time, version number rules are defined to implement version tracking of model code and parameter files based on Git, supporting fast rollback of any historical version within 15 minutes.

2. The integration and maintenance method for a secondary model server for steel rolling as described in claim 1, characterized in that: In step S11, virtualization and containerization deployment specifically include: (1) For the roll gap verification model, physical server resources are pooled through VMware vSphere or KVM virtualization platform, with a single physical machine carrying ≥8 virtual machines; (2) The roll wear prediction model is encapsulated in Docker containers and managed by Kubernetes clusters to control cross-container communication latency at the 10ms level.

3. The integration and maintenance method for a secondary model server for steel rolling as described in claim 1, characterized in that: In step S12, the hybrid cloud architecture design specifically includes: (1) Deploy the rolling process calculation model using Nutanix to ensure local data interaction latency < 5ms; (2) Model training and historical data archiving are migrated to the enterprise private cloud or hybrid cloud to accelerate model iteration using cloud GPU resources.

4. The integration and maintenance method for a secondary model server for steel rolling as described in claim 1, characterized in that: In step S21, the server hardware indicators include CPU, memory, and disk I / O; the model running status includes computation time and convergence rate; the unified monitoring module is Prometheus+Grafana; the dynamic threshold alarm is based on the 3σ rule; and a level 1 alarm, i.e., model crash, will trigger a switch to a backup model within 500ms.

5. The integration and maintenance method for a secondary model server for steel rolling as described in claim 1, characterized in that: In step S22, Terraform is responsible for the Infrastructure as Code (IaC), Ansible is responsible for the Configuration as Code (MAC), and the rolling force calculation error is ≤ ±1%.

6. The integration and maintenance method for a secondary model server for steel rolling as described in claim 1, characterized in that: In step S31, each logic process group is assigned a responsible maintenance personnel, and a cross-group joint debugging meeting is established, with the frequency controlled at once per week.

7. The integration and maintenance method for a secondary model server for steel rolling as described in claim 1, characterized in that: In step S32, cross-process data retrieval includes calling hot-rolled roll wear data from the cold-rolled sheet shape model.

8. The integration and maintenance method for a secondary model server for steel rolling according to any one of claims 1 to 7, characterized in that: In steps S1 to S3, the virtualization platform VMware / KVM, the container cluster Docker / Kubernetes, the hyperconverged architecture Nutanix, and cloud resources constitute the hardware layer.

9. The integration and maintenance method for a secondary model server for steel rolling according to any one of claims 1 to 7, characterized in that: In steps S1 to S3, the management layer consists of the unified monitoring module Prometheus+Grafana, the configuration management module CMDB+IaC, and the automated operation and maintenance module.

10. The integration and maintenance method for a secondary model server for steel rolling according to any one of claims 1 to 7, characterized in that: In steps S1 to S3, the Model Service Bus (MSB) based on process domain grouping forms the application layer, which supports API calls and version management across process models and connects to the steel rolling production control system.