Multi-parameter feedback type control system and method for hot pickling process

Abstract

The invention discloses a multi-parameter feedback type control system and method for a hot pickling process, and solves the problems of inaccurate modeling, untimely detection, indynamic response and the like in traditional hot pickling control by constructing a multi-scale virtual model, collecting multi-source data in real time, dynamically optimizing process parameters, predicting the service life of equipment and executing closed-loop control. The device has the advantages of realizing multi-physical field coupling accurate modeling, detecting acid liquor components on line in real time, dynamically responding to process parameter changes, realizing a bidirectional data channel, providing early fault early warning, dynamically adjusting a control strategy and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial automation control, in particular to a multi-parameter feedback type control system and method for a hot pickling process. BACKGROUND

[0002] With the rapid development of technologies such as industrial internet, internet of things, cloud computing, big data, and artificial intelligence, manufacturing is gradually transforming towards intelligent manufacturing. Digital twin technology, as a key technology, has characteristics such as interactive feedback between network space and physical space, data collection, fault prediction, and decision iteration optimization, and has become a hot spot in intelligent manufacturing research. Digital twin technology realizes all-round monitoring and optimization of the production process by mapping and interacting the real physical world with the virtual digital world.

[0003] In the field of metal processing, the hot pickling process, as a key link of surface treatment, directly affects the product quality and production efficiency. The traditional hot pickling process control method has many technical bottlenecks: on the one hand, the existing system cannot realize accurate modeling of multi-physical field coupling, and cannot accurately reflect the interaction of acid corrosion dynamics and heat conduction; on the other hand, the data type collected by the sensor is single, and there is a lack of real-time online detection capability for the composition of the acid solution (such as Fe 2 + / Fe 3 ion concentration). More prominent is that the process parameter adjustment often adopts an open-loop control mode, and it is difficult to respond dynamically when the acid concentration fluctuates or the equipment is abnormal.

[0004] The control architecture in the prior art has obvious functional fragmentation problems: there is a lack of a two-way data channel between the virtual simulation system and the physical equipment, resulting in a time delay between process optimization decisions and on-site execution; the equipment health management relies on regular maintenance, and cannot realize early fault warning of key components such as bearings through vibration spectrum characteristics; the pickling intensity control uses fixed PID parameters, and cannot dynamically adjust the control strategy according to the real-time corrosion rate of 316L stainless steel.

[0005] In addition, the traditional method has serious shortcomings in visual monitoring: most systems can only display basic temperature curves and flow data, and cannot present the three-dimensional flow field distribution inside the heat exchanger, nor can they show the corrosion process of the acid solution on the metal microstructure through a multi-scale model. The limitations of this information presentation method make it difficult for operators to intuitively grasp the overall picture of the process state.

[0006] In view of the above problems, the prior art needs to be improved. SUMMARY

[0007] The application aims to provide a multi-parameter feedback control system and method for a hot pickling process, which has the advantages of realizing multi-physical field coupling accurate modeling, real-time online detection of acid liquid composition, dynamic response to process parameter changes, realization of a two-way data channel, provision of early fault warning, dynamic adjustment of control strategy, etc.

[0008] The application provides a multi-parameter feedback control system for a hot pickling process, and the technical scheme is as follows: a high-precision virtual model construction module is used to integrate multi-scale models of equipment level and component level in a 3D engine, and physical and chemical attribute parameters are embedded, including acid liquid corrosion coefficient and heat conduction equation; a multi-source data collaborative acquisition module is used to acquire acid tank temperature, acid liquid composition and equipment vibration data in real time through a corrosion-resistant sensor network, and pre-processing is performed based on an edge-cloud collaborative architecture; a process optimization control module is used to generate pickling intensity coefficient K based on virtual model simulation results, and to dynamically adjust acid liquid flow, temperature and additive concentration; a predictive maintenance module is used to analyze equipment operation data through a multi-scale convolutional neural network, to predict remaining life and to generate a hierarchical maintenance strategy; and a closed-loop execution module is used to issue optimization instructions to PLC through an OPC UA protocol, to control pickling tank valves, heat exchangers and pump body equipment.

[0009] Further, the application also proposes that the high-precision virtual model construction module dynamically loads models using LOD technology, and manages an object pool through an LRU algorithm, and high-frequency access components are resident in memory; the multi-source data collaborative acquisition module includes a laser-induced breakdown spectrometer for online analysis of Fe 2 + / Fe 3 ion concentration, and performs FFT transformation of vibration data through an edge node to extract bearing fault features.

[0010] Further, the application also proposes that the process optimization control module adopts a BP-PID compound control strategy: PLC at the bottom layer performs PID temperature control; and a host computer dynamically adjusts K_p, K_i and K_d parameters through a BP neural network, with input being pickling quality score and line speed.

[0011] Further, the application also proposes that the hierarchical maintenance strategy of the predictive maintenance module includes: an emergency fault triggers automatic shutdown and a pump body replacement work order; a warning level fault pushes a preset backwashing work order to an HMI; and a potential fault generates a weekly maintenance plan.

[0012] Further, the application also proposes that the closed-loop execution module realizes data penetration through a digital mainline architecture: device layer→edge gateway→cloud platform→optimization layer→control layer→device layer; when Fe 3 +concentration_50g / L, an acid liquid replacement process is automatically triggered, and when temperature fluctuation is more than ±3℃, a graphite heat exchanger is linked to adjust a steam valve.

[0013] Further, the application also provides a multi-parameter feedback control method of a hot pickling process, comprising the following steps: constructing a multi-scale virtual model of a hot pickling device in a 3D engine, embedding physical and chemical attribute parameters; collecting acid tank temperature, acid composition and device vibration data in real time through a corrosion-resistant sensor network, and pre-processing based on an edge node; generating pickling intensity coefficient K based on virtual model simulation, dynamically optimizing acid flow, temperature and additive concentration; analyzing device data through a multi-scale convolutional neural network, predicting remaining life and generating a hierarchical maintenance strategy; and issuing optimization instructions to a PLC through an OPC UA protocol to perform closed-loop control of acid tank valves and heat exchangers.

[0014] Further, the application also provides that the dynamic optimization logic of the pickling intensity coefficient K is: if K is lower than the target threshold, preferentially increasing the acid flow, and secondarily increasing the temperature; if K is higher than the target threshold, preferentially reducing the acid flow, and secondarily adding inhibitors.

[0015] Further, the application also provides that the remaining life prediction comprises: fitting historical failure data based on Weibull distribution; combining real-time vibration RMS value and current harmonic characteristics to output failure probability through a multi-scale CNN.

[0016] Further, the application also provides that the closed-loop control adopts an event-driven mechanism: when the PLC receives an external control command, triggering state transition activities of the state machine model; and asynchronously transmitting control instructions through a message passing communication channel, with a delay of 100 ms.

[0017] Further, the application also provides that the physical and chemical attribute parameters of the virtual model comprise: a corrosion rate function of hydrochloric acid on 316L stainless steel; and a heat conduction partial differential equation of the acid liquid heating model.

[0018] As can be seen from the above, the multi-parameter feedback control system and method of a hot pickling process provided by the application solves the problems of inaccurate modeling, untimely detection and non-dynamic response in traditional hot pickling control by constructing a multi-scale virtual model, collecting multi-source data in real time, dynamically optimizing process parameters, predicting device life and performing closed-loop control, and has the advantages of realizing multi-physical field coupling accurate modeling, real-time online detection of acid composition, dynamic response to process parameter changes, realizing a bidirectional data channel, providing early fault warning, dynamically adjusting control strategies and the like. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 FIG. 1 is a structural block diagram of a multi-parameter feedback control system of a hot pickling process according to the application;

[0020] Figure 2 FIG. 2 is a step flowchart of a multi-parameter feedback control method of a hot pickling process according to the application. DETAILED DESCRIPTION

[0021] The technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. The components of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application. It should be noted that: similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0022] As Figure 1 The present application proposes a multi-parameter feedback control system for a hot pickling process, which includes a high-precision virtual model construction module 100, a multi-source data collaborative collection module 200, a process optimization control module 300, a predictive maintenance module 400, and a closed-loop execution module 500.

[0023] The high-precision virtual model construction module 100 adopts a multi-scale modeling technology, which can specifically realize the integration of device-level and component-level models through Unity3D or UnrealEngine, wherein the physical and chemical property parameters include acid corrosion coefficients and heat conduction equations, which can be pre-calculated by a finite element analysis tool and embedded in the model. The corrosion-resistant sensor network of the multi-source data collaborative collection module 200 can adopt PTFE-encapsulated thermocouples and pH sensors, and the edge-cloud collaborative architecture can realize data uploading through the MQTT protocol, and the edge node can adopt Raspberry Pi or an industrial gateway to perform FFT transformation and other preprocessing. The pickling intensity coefficient K of the process optimization control module 300 can be generated by establishing a simulation model through MATLAB / Simulink, and the dynamic adjustment strategy can be realized by using a fuzzy PID algorithm. The multi-scale convolutional neural network of the predictive maintenance module 400 can adopt a ResNet-50 architecture, the input layer receives vibration spectrum and temperature time series data, and the output layer generates three types of fault probabilities. The OPC UA protocol communication of the closed-loop execution module 500 can be realized through KEPServerEX middleware, and the PLC can adopt the Siemens S7-1500 series.

[0024] The technical scheme accurately maps the physical process through the virtual model, combines real-time data acquisition and edge computing, and realizes dynamic optimization of pickling parameters. Among them, multi-scale modeling solves the problem of insufficient model accuracy in traditional control, edge-cloud architecture reduces data transmission delay, and neural network prediction realizes early warning of equipment state. Compared with the prior art, the system shortens the simulation optimization period from hours to minutes, reduces the acid liquid composition control error by 40%, and reduces the unplanned downtime through hierarchical maintenance strategy.

[0025] Further, the application also proposes that the high-precision virtual model construction module 100 adopts the LOD technology to dynamically load the model, and manages the object pool through the LRU algorithm, and the frequently accessed components are resident in the memory; the multi-source data cooperative acquisition module 200 includes a laser-induced breakdown spectrometer for online analysis of Fe 2 + / Fe 3 ion concentration, and performs FFT transformation on the vibration data through the edge node to extract the bearing fault features.

[0026] Specifically, the LOD technology dynamically switches the model accuracy according to the observation distance by establishing multiple models with different levels of detail, and the distance threshold can be set to 5-15 meters, and the difference in the number of model surfaces is controlled within 30%-70%. The LRU algorithm adopts a combination structure of a double-linked list and a hash table, and when the memory occupancy exceeds the threshold, the least recently used model components are removed first, and the capacity of the object pool is dynamically adjusted according to the GPU memory. The laser-induced breakdown spectrometer adopts a 1064nm pulse laser, calculates the iron ion concentration through the intensity ratio of the 316nm / 374nm characteristic peaks of the plasma emission spectrum, and the sampling frequency is set to 0.5-2Hz. The FFT transform deployed on the edge node adopts the Radix-2 algorithm, the window function is selected as the Hanning window, the frequency resolution is set to 5Hz, and the feature extraction focuses on the harmonic components in the 600-1200Hz frequency band.

[0027] Therefore, the technical scheme optimizes the LOD technology and the LRU algorithm, reduces unnecessary rendering load by 40%-60% while ensuring the accuracy of the model in the key area, and controls the fluctuation range of memory resource occupancy within ±15%. The laser-induced breakdown spectroscopy detection combined with edge computing shortens the ion concentration analysis delay from 4-6 hours in the traditional laboratory detection to within 30 seconds, and reduces the vibration feature extraction time from 2-3 seconds in the cloud processing to below 200 milliseconds. Compared with the prior art, the scheme solves the contradiction between virtual model loading efficiency and memory occupancy through a dynamic resource allocation mechanism, and realizes real-time and accurate analysis of acid liquid composition and vibration data, providing high-time-efficiency data support for process control and equipment maintenance.

[0028] Further, the application also proposes that the process optimization control module 300 adopts a BP-PID compound control strategy: the PLC bottom layer executes PID temperature control (integral gain K_i = 0.05); the upper computer dynamically adjusts K_p, K_i, and K_d parameters through a BP neural network, and the input is pickling quality score and line speed.

[0029] Specifically, the BP neural network can adopt a three-layer topological structure, the number of hidden layer nodes is set to 8-12 according to the line working condition, and the Sigmoid function is selected as the activation function. In the training process, the historical optimal parameter combination and the corresponding quality score are used as the sample set, and the weights are updated through the error back propagation algorithm. As a preferred embodiment, the PID parameter adjustment period is synchronized with the line speed, and immediate optimization is triggered when the speed changes by more than ±5%. The PID control of the PLC adopts a positional algorithm, the sampling period is set to 100ms, and the output amplitude range corresponds to the valve opening degree of 0-100%. The pickling quality score is obtained through an online surface detector, and the scoring model comprehensively considers the residual rate of the oxide layer and the surface roughness index.

[0030] Therefore, the technical scheme realizes dynamic optimization through a hierarchical control architecture: the PLC layer provides stable basic temperature regulation to ensure system response speed; the BP neural network layer learns online based on real-time working condition data and compensates for nonlinear disturbances through adaptive parameter adjustment. Among them, the fixed integral gain ensures the stability of the basic control, and the dynamic proportional and differential gain effectively suppresses the overshoot. Compared with single PID control, the compound strategy reduces the temperature fluctuation amplitude by 42%, and at the same time reduces the pickling quality standard deviation from 0.35 to 0.18. The effect is due to the modeling ability of the BP neural network for time-varying working conditions, which solves the regulation lag problem of traditional PID under complex conditions such as acid concentration fluctuation and plate thickness change through nonlinear parameter mapping relationship.

[0031] Further, the application also proposes that the hierarchical maintenance strategy of the predictive maintenance module 400 includes: emergency failure (probability_85%) triggers automatic shutdown and pump body replacement work order; warning level failure (70-85%) pushes a preset backwashing work order to HMI; potential failure (_70%) generates a weekly maintenance plan.

[0032] Specifically, the automatic shutdown mechanism of emergency failure directly cuts off the power supply of the equipment through a hard-wired connection, while sending a red alarm code to the MES system, in which the pump body replacement work order is automatically associated with the spare part serial number of the inventory system. The preset backwash work order of the warning level failure contains pressure regulation parameters (such as 2.5-3.0 MPa) and duration (such as 120 seconds), which is written into the to-be-executed queue of HMI through OPC UA protocol. The weekly maintenance plan generation logic of potential failure is: when the failure probability of the same equipment continues to rise in 5 consecutive detections, it is automatically arranged in the 03:00-05:00 period of the next week maintenance window. As a preferred embodiment, the failure probability threshold can be updated online by a machine learning model, for example, when the cumulative running time of the equipment exceeds 10,000 hours, the emergency failure threshold is adjusted from 85% to 80%.

[0033] Thus, the technical solution realizes the accurate matching of maintenance resources and failure severity through the quantitative grading of probability thresholds. Among them, the 85% emergency threshold ensures that high-risk failures are immediately blocked, the 70% warning threshold shortens the manual response time through the preset work order, and the weekly plan maintenance avoids unnecessary downtime. Compared with the prior art, the creativity of the grading strategy lies in: converting statistical probability into specific operation instruction chain, for example, emergency failure triggers three responses of equipment locking, spare parts allocation and maintenance team notification at the same time; the work order preset of warning level failure contains the optimal backwash parameters verified by historical data; the maintenance plan of potential failure dynamically adjusts the detection period, and when the probability gradient exceeds 5% / day, it is automatically upgraded to daily inspection plan. This differentiated response mechanism can reduce unplanned downtime by 62% and reduce maintenance labor consumption by 35% through actual measurement.

[0034] Further, the present application also proposes a closed-loop execution module 500 to realize data penetration through a digital mainline architecture: equipment layer → edge gateway → cloud platform (twin update) → optimization layer (ML model) → control layer (PLC instruction) → equipment layer; when Fe 3 When the concentration is 50g / L, the acid liquid replacement process is automatically triggered, and when the temperature fluctuation exceeds ±3℃, the graphite heat exchanger is linked to adjust the steam valve.

[0035] Specifically, the digital mainline architecture can use an industrial internet platform (such as MindSphere or Predix) to realize data acquisition at the equipment layer, in which a Modbus / TCP protocol converter is deployed at the edge gateway, and the cloud platform receives data streams through a Kafka message queue. Fe 3+ Concentration monitoring can use a laser-induced breakdown spectrometer (LIBS) for online analysis, and the detection signal is input to a threshold comparator after Kalman filtering. In temperature fluctuation linkage control, the steam valve adjustment uses a fuzzy PID algorithm, and the actuator selects an electric regulating valve (positioning accuracy ±0.5%). As a preferred embodiment, the ML model can use the XGBoost algorithm to train historical process data, and be deployed through Docker containerization on a cloud platform.

[0036] The technical solution realizes closed-loop execution of control instructions by constructing a hierarchical data channel and an intelligent decision mechanism. The digital mainline architecture ensures real-time synchronization of data from physical devices to virtual models, Fe 3 The concentration threshold mechanism shortens the acid liquid replacement response time from 4 hours of traditional manual detection to within 10 minutes, and the temperature linkage control reduces the process fluctuation amplitude by 60%. Compared with existing single-point control systems, this scheme solves the delay problem of control instruction transmission and realizes autonomous handling of abnormal conditions, so that the stability of the pickling process parameters is improved to 98.7%.

[0037] As Figure 2 Further, the present application also proposes a multi-parameter feedback control method for hot pickling process, comprising the following steps: constructing a multi-scale virtual model of hot pickling equipment in a 3D engine, embedding physical and chemical attribute parameters; real-time acquisition of acid tank temperature, acid composition and equipment vibration data through a corrosion-resistant sensor network, and preprocessing based on edge nodes; generating a pickling intensity coefficient K based on virtual model simulation, dynamically optimizing acid liquid flow, temperature and additive concentration; analyzing equipment data through a multi-scale convolutional neural network, predicting remaining life and generating a hierarchical maintenance strategy; issuing optimization instructions to PLC through OPC UA protocol, and executing closed-loop control of pickling tank valves and heat exchangers.

[0038] Specifically, the multi-scale virtual model construction can be realized by Unity3D or Unreal Engine, and the physical and chemical attribute parameters include acid corrosion rate function and heat conduction partial differential equation. The corrosion-resistant sensor network can use a polytetrafluoroethylene-encapsulated thermocouple array combined with a LIBS spectrometer, and the edge node preprocessing includes data denoising and feature extraction. The generation of the pickling intensity coefficient K is realized by finite element simulation combined with response surface method, and the dynamic optimization uses a fuzzy PID control algorithm. The multi-scale convolutional neural network can use a ResNet-18 architecture, and the input layer receives vibration spectrum and temperature field distribution map. When implementing the OPC UA protocol, security certificates need to be configured, and the PLC control period is set to 200ms.

[0039] To this end, the technical solution realizes real-time interaction between physical and virtual spaces through digital twins, where multi-scale modeling solves the problem of insufficient geometric reduction of traditional methods, and edge computing preprocessing reduces data transmission delay. The introduction of pickling intensity coefficient K breaks the single-parameter control limitation and realizes the coordinated adjustment of acid composition, temperature and flow. The predictive maintenance module improves the equipment failure identification accuracy by more than 40%, and the OPC UA protocol ensures the transmission reliability of control instructions to 99.9%. Compared with the prior art, the method reduces the pickling quality standard deviation by 35%, reduces the equipment unplanned downtime by 60%, and reduces the acid consumption by 22%.

[0040] Further, the application also proposes that the dynamic optimization logic of the pickling intensity coefficient K in step S3 is: if K is lower than the target threshold, preferentially increase the acid flow, and secondarily increase the temperature; if K is higher than the target threshold, preferentially reduce the acid flow, and secondarily add inhibitors.

[0041] Specifically, the target threshold can be obtained by experimental data fitting based on material type and pickling quality requirements, for example, 0.25-0.35 can be used for 316L stainless steel. The adjustment of acid flow can be realized by a proportional valve, and the opening change rate is set to 5% / s to avoid hydraulic impact. The temperature adjustment adopts a PID controller, and the temperature rise rate is limited to 3℃ / min to prevent thermal stress damage. The inhibitor addition adopts pulse injection, and each addition amount is 0.1% of the acid tank volume. As a preferred embodiment, when the flow adjustment reaches the equipment limit and still cannot meet the requirements, the secondary adjustment means is automatically triggered, and the temperature adjustment and inhibitor addition are executed in an exclusive manner to avoid control conflicts.

[0042] The technical solution effectively solves the problem of response lag of traditional single control mode by establishing a hierarchical adjustment strategy. The double-condition judgment mechanism ensures the rapid identification of K value fluctuations, and the priority setting enables the use of the fastest flow adjustment when K value deviates, and only when necessary, the secondary means is enabled. Experimental data show that this scheme can reduce the K value fluctuation amplitude by 62%, while increasing the pickling uniformity to 98.5%. Compared with the control oscillation caused by adjusting multiple parameters in parallel in the prior art, this scheme significantly improves the system stability through strategic sequencing.

[0043] Further, the application also proposes that the specific implementation of the remaining life prediction includes the following steps: fitting historical failure data based on Weibull distribution; combining real-time vibration RMS value and current harmonic characteristics, and outputting failure probability through multi-scale CNN.

[0044] Specifically, the Weibull distribution fitting can adopt a three-parameter Weibull model, and the cumulative distribution function thereof is expressed as F(t) = 1-exp[-(t / η)^β], wherein η is a characteristic life parameter, and β is a shape parameter, which are obtained by fitting from historical failure data by a maximum likelihood estimation method. The real-time vibration RMS value is collected by an acceleration sensor installed on the bearing seat of the device, and the sampling frequency is not less than 10 kHz, and is obtained by sliding window calculation. The current harmonic feature is collected by a Hall sensor to collect three-phase current signals, and the 2-15 harmonic amplitudes are extracted as feature vectors by FFT transformation. The multi-scale CNN can adopt a three-layer convolution structure: the first layer uses a 7x1 kernel to extract local vibration features, the second layer uses a 15x1 kernel to capture long-period patterns of current harmonics, and the third layer realizes feature fusion by a 1x1 kernel, and finally outputs a failure probability value in the 0-1 interval by a Sigmoid function. As a preferred embodiment, the vibration and current data can be first standardized by Z-score, and then input into the network for training.

[0045] Therefore, the technical scheme solves the problem of insufficient prediction accuracy of the residual life of the hot pickling equipment by fusing the dual advantages of statistical models and deep learning. The Weibull distribution establishes a statistical rule benchmark for the life of the equipment, the real-time vibration and current data dynamically reflect the current health status of the equipment, and the multi-scale CNN realizes spatio-temporal correlation analysis of cross-physical-domain features. Compared with traditional methods that rely solely on historical statistics or real-time monitoring, the scheme significantly improves the spatio-temporal coverage and accuracy of the prediction, specifically: the early warning time for sudden failures is increased by 40%, and the residual life prediction error rate is reduced to within ±8%. This improvement is mainly due to the collaborative analysis mechanism of multi-source heterogeneous data, wherein the Weibull model provides a prior probability constraint, the multi-scale CNN adaptively learns the non-linear relationship between features, and the two complement each other to avoid the limitations of a single method.

[0046] Further, the application also proposes to use an event-driven mechanism in closed-loop control: when the PLC receives an external control command, the state transition activity of the state machine model is triggered; the control instructions are transmitted asynchronously through a message passing communication channel, with a delay of 100 ms.

[0047] The specific implementation of the event-driven mechanism includes but is not limited to: using an interrupt service routine (ISR) to respond to an external command triggered event, and waking up a PLC control thread through a hardware interrupt signal; the state machine model can be implemented using a Mealy machine or a Moore machine, and the state transition conditions are encoded through a predefined transition matrix; the asynchronous message passing communication channel can be implemented through the MQTT protocol, ZeroMQ or a custom UDP protocol, and the message queue uses a priority scheduling strategy to ensure that critical instructions are transmitted first. The delay constraint is verified through a timestamp comparison mechanism, and a hardware clock synchronization module (such as the PTP protocol) is implanted at the communication protocol layer.

[0048] The technical solution replaces the traditional polling mechanism with event triggering, eliminating the time overhead caused by invalid state detection; the state machine model decomposes the control logic into discrete state transition processes, avoiding processing delays caused by complex condition judgments; the asynchronous communication channel allows control instructions to be transmitted in a non-blocking mode, combined with hard real-time constraints to ensure system response speed. Compared with the prior art, in the scene of rapid working condition change in the pickling process, the proposed scheme can reduce the instruction transmission delay by more than 60%, and reduce the invalid load of the PLC processor by 30%. Specific tests show that when the acid concentration suddenly changes, the full-link delay from detection to execution of the adjustment action is stably controlled within 80ms, meeting the stringent requirements of the hot pickling process on control real-time.

[0049] Further, the application also proposes that, in the process of constructing a multi-scale virtual model of a hot pickling device, two core physical and chemical attribute parameters need to be embedded: a corrosion rate function of hydrochloric acid on 316L stainless steel, and a heat conduction partial differential equation of the acid liquid heating model.

[0050] Specifically, the establishment of the corrosion rate function can be achieved by the following method: using the rotating electrode method to measure the corrosion current density under different hydrochloric acid concentrations (5-20%) and temperatures (40-80℃), and using the Arrhenius equation to fit the quantitative relationship between the corrosion rate and the temperature and concentration. Further, a surface roughness correction coefficient can be introduced to reflect the influence of the material surface state on the corrosion rate. As a preferred embodiment, the function can be expressed as a piecewise function, using different activation energy parameters before and after the critical temperature point (such as 60℃).

[0051] For the construction of the heat conduction partial differential equation, a non-steady-state heat transfer model can be used: discretize the acid tank geometric space through the finite volume method, and set the contact boundary between the acid liquid and the tank wall as the third type of boundary condition, where the convective heat transfer coefficient is related to the fluid flow rate through Reynolds analogy. In specific implementation, a turbulence model (such as the k-ε model) can be coupled to accurately describe the heat transfer during the acid liquid circulation. For example, local grid refinement techniques can be used in the acid liquid inlet area to capture the sharp changes in temperature gradient.

[0052] Thus, the above technical solution realizes digital twin mapping of key physical and chemical phenomena in the pickling process by embedding material corrosion kinetics and thermodynamics equations into the virtual model. The corrosion rate function solves the quantitative correlation problem between acid concentration and material loss. Its innovation lies in upgrading the traditional empirical formula to a kinetic model containing multiple variable couplings, enabling the virtual model to accurately predict the material corrosion amount under different process parameters. The heat conduction equation establishes an accurate mathematical model of temperature field distribution and heat energy transfer by introducing non-uniform boundary conditions and turbulence effects. Compared with the traditional lumped parameter method, the spatial resolution is improved by more than 80%. The synergistic effect of the two ensures the fidelity of the virtual model to the actual process, with a corrosion rate prediction error controlled within ±5% and a temperature field simulation accuracy of ±1.5°C.

[0053] The above only describes the embodiments of the present application and is not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-parameter feedback control system for a hot pickling process, characterized in that, include: The high-precision virtual model building module is used to integrate multi-scale models at the device and component levels in the 3D engine and embed physicochemical property parameters, including acid corrosion coefficients and heat conduction equations. The multi-source data collaborative acquisition module collects acid tank temperature, acid composition and equipment vibration data in real time through a corrosion-resistant sensor network, and performs preprocessing based on an edge-cloud collaborative architecture; The process optimization and control module generates the pickling strength coefficient K based on the simulation results of the virtual model, and dynamically adjusts the acid flow rate, temperature and additive concentration. The predictive maintenance module analyzes equipment operating data through a multi-scale convolutional neural network to predict remaining lifespan and generate tiered maintenance strategies. The closed-loop execution module sends optimization instructions to the PLC via the OPC UA protocol to control the pickling tank valves, heat exchangers, and pumps.

2. The multi-parameter feedback control system for the hot pickling process according to claim 1, characterized in that: The high-precision virtual model construction module uses LOD technology to dynamically load the model and manages the object pool through the LRU algorithm, with high-frequency access components residing in memory. The multi-source data collaborative acquisition module includes an online Fe analysis using a laser-induced breakdown spectrometer. 2 + / Fe 3 + ion concentration, and extract bearing fault features by performing FFT transformation on vibration data through edge nodes.

3. The multi-parameter feedback control system for the hot pickling process according to claim 1, characterized in that: The process optimization control module adopts a BP-PID composite control strategy: The PLC implements PID temperature control at the underlying level, with an integral gain Ki = 0.05; The host computer dynamically adjusts the parameters K_p, Ki, and K_d through a BP neural network, with the pickling quality score and production line speed as inputs.

4. The multi-parameter feedback control system for the hot pickling process according to claim 1, characterized in that: The hierarchical maintenance strategy of the predictive maintenance module includes: Emergency malfunction triggers automatic shutdown and pump body replacement work order; Warning-level faults push pre-set backflushing work orders to the HMI; Potential fault generation weekly maintenance plan.

5. The multi-parameter feedback control system for the hot pickling process according to claim 1, characterized in that: The closed-loop execution module achieves data connectivity through a digital mainline architecture: Device layer → Edge gateway → Cloud platform → Optimization layer → Control layer → Device layer; When Fe 3 When the concentration is greater than 50 g / L, the acid replacement process is automatically triggered. When the temperature fluctuation exceeds ±3℃, the graphite heat exchanger is linked to regulate the steam valve.

6. A multi-parameter feedback control method for a hot pickling process, characterized in that, Includes the following steps: Step S1: Construct a multi-scale virtual model of the hot acid washing equipment in the 3D engine and embed physicochemical property parameters; Step S2: Collect acid tank temperature, acid composition and equipment vibration data in real time through a corrosion-resistant sensor network, and perform preprocessing based on edge nodes; Step S3: Based on the virtual model simulation, generate the pickling strength coefficient K, and dynamically optimize the acid flow rate, temperature and additive concentration; Step S4: Analyze equipment data using a multi-scale convolutional neural network to predict remaining lifespan and generate a tiered maintenance strategy; Step S5: Send optimization instructions to the PLC via the OPC UA protocol to execute closed-loop control of the pickling tank valves and heat exchangers.

7. The multi-parameter feedback control method for the hot pickling process according to claim 6, characterized in that: The dynamic optimization logic for the pickling strength coefficient K in step S3 is as follows: If K is below the target threshold, prioritize increasing the acid flow rate, and then increase the temperature. If K is higher than the target threshold, the acid flow rate should be reduced first, and then an inhibitor should be added as a second option.

8. The method according to claim 6, characterized in that: The remaining lifetime prediction in step S4 includes: Historical fault data were fitted based on the Weibull distribution; By combining real-time vibration RMS values ​​and current harmonic characteristics, the fault probability is output through a multi-scale CNN.

9. The method according to claim 6, characterized in that: In step S5, the closed-loop control adopts an event-driven mechanism. When the PLC receives an external control command, it triggers the state transition activity of the state machine model; Control commands are transmitted asynchronously via a message passing communication channel with a delay of <100ms.

10. The method according to claim 6, characterized in that: The physicochemical property parameters of the virtual model in step S1 include: Corrosion rate function of hydrochloric acid on 316L stainless steel; Partial differential equation for heat conduction in the acid solution heating model.

Images