VOCs treatment equipment fault diagnosis method and system based on digital twinning

Abstract

The application provides a VOCs treatment equipment fault diagnosis method and system based on digital twinning, relates to the field of industrial VOCs waste gas treatment, and comprises the following steps: constructing a VOCs equipment global multi-physical field digital twinning body integrating a fluid field, a thermal field, an electric control signal and catalytic reaction kinetics; establishing a bidirectional data interaction link between the entity equipment and the twinning body; and synchronously acquiring real-time operation data of the entity equipment, waste gas emission monitoring data and twinning body fault-free standard simulation data. The application couples finite volume method, finite element transient solution algorithm, Kalman time series filtering algorithm and Arrhenius kinetics fitting algorithm to build a fluid field, a thermal field, an electric control signal, catalytic reaction kinetics and multi-physical field global twinning model, simultaneously uses IEEE1588v2 precise clock synchronization protocol and sliding time series alignment algorithm to complete global data time series calibration, and realizes complete copying of multi-dimensional operation conditions of the equipment and high-precision synchronization of virtual and real data.

Description

Technical Field

[0001] This invention relates to the field of industrial VOCs waste gas treatment, specifically to a method and system for fault diagnosis of VOCs treatment equipment based on digital twins. Background Technology

[0002] In the field of industrial VOCs waste gas treatment, the development of equipment fault diagnosis technology has evolved from manual experience-based inspections and single-parameter threshold alarms to basic digital twin status monitoring. Early VOCs catalytic treatment equipment relied on manual on-site inspections by maintenance personnel, who judged equipment malfunctions based on experience. This resulted in significant monitoring delays and an inability to identify hidden operational anomalies. Subsequently, the industry generally adopted single-parameter threshold alarms, setting fixed thresholds for key parameters such as temperature, air pressure, and waste gas concentration. Exceeding these limits resulted in equipment malfunctions, achieving automated online monitoring. In recent years, digital twins have been gradually applied to the operation and maintenance of environmental protection equipment. By building simple simulation models to replicate the operating status of physical equipment, and relying on the discrepancy between virtual and real data, preliminary anomaly detection is achieved, further improving the intelligence level of equipment monitoring.

[0003] Although existing technologies can monitor basic operational anomalies in VOCs treatment equipment, the following problems still exist in actual industrial environments with multiple interferences, coupled linkages of multiple components, and long-term aging of equipment: 1. Existing digital twin models lack complete physical field coverage and have insufficient accuracy in synchronizing virtual and real data timing. Most existing conventional twin models only simulate and model a single physical field, failing to reproduce the true coupled operating mechanism of the equipment; 2. The fault monitoring methods are simplistic and cannot filter out operational interference, and the capabilities for fault classification and environmental risk assessment are lacking. Existing technologies only employ passive, single residual monitoring methods, which cannot distinguish between invalid interference signals caused by on-site environmental fluctuations and production load switching and genuine equipment fault signals, resulting in high false alarm and false alarm rates. Furthermore, they cannot accurately differentiate between three types of faults: single-component hard faults, multi-component coupled soft faults, and process parameter drift-induced latent faults. They also cannot combine VOCs emission offsets to complete environmental risk classification, resulting in coarse fault diagnosis results that cannot guide precise on-site operation and maintenance.

[0004] 3. The twin model lacks closed-loop correction and autonomous iteration mechanisms, resulting in continuous accumulation of errors over long-term operation.

[0005] Therefore, a fault diagnosis method and system for VOCs treatment equipment based on digital twins is needed to solve the above problems. Summary of the Invention

[0006] Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for fault diagnosis of VOCs treatment equipment based on digital twins, which solves the problems mentioned in the background section.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a fault diagnosis method for VOCs treatment equipment based on digital twins, comprising the following steps: S1. Construct a full-domain multi-physics digital twin of VOCs equipment that integrates fluid field, thermal field, electronic control signal and catalytic reaction dynamics, establish a two-way data interaction link between the physical equipment and the twin, and simultaneously acquire real-time operating data of the physical equipment, exhaust emission monitoring data and fault-free standard simulation data of the twin. S2. Construct a dual-parallel diagnostic architecture combining passive residual monitoring and active micro-disturbance. The architecture consists of a normal monitoring branch and an abnormal disturbance branch that operate in parallel. The normal monitoring branch calculates the residuals of the virtual and real operating parameters in real time, separates the residuals of environmental noise and non-fault interference caused by operating condition switching, and obtains normal static residual data. When the normal static residual data does not exceed the safety threshold, the normal monitoring branch continues to operate, determines that the equipment is currently in a normal state or an acceptable slight drift state, and continuously updates the normal static residual data. When the normal static residual data exceeds the safety threshold, the abnormal disturbance branch is triggered, and a controllable small-amplitude micro-disturbance signal is sent by the twin to synchronously act on the physical equipment to obtain the dynamic response data of the disturbance. S3. By combining normal static residual data and disturbance dynamic response data, the directed spectrum of equipment fault propagation is matched to distinguish between single component hard faults, multi-component coupled soft faults and process parameter drift hidden faults. The environmental impact level of the fault is determined by synchronously associating the VOCs emission concentration offset. S4. Conduct virtual-real two-way closed-loop correction simulation based on twins to eliminate the inherent modeling errors of twin models; S5 outputs accurate fault location results, fault environmental risk level, non-stop emergency control plan and shutdown maintenance plan. At the same time, it puts disturbance dynamic response data, fault-related emission data and model correction data into the database in a closed loop and updates the fault map and twin simulation model autonomously.

[0009] Preferably, the specific steps of S1 are as follows: S1.1 The fluid field mesh modeling is completed using the finite volume method, the thermal transient temperature field mapping modeling is completed using the finite element transient solution algorithm, the electronic control signal time synchronization modeling is completed using the Kalman time-series filtering algorithm, and the catalytic reaction kinetic parameter fitting modeling is completed using the Arrhenius dynamics fitting algorithm. The four independent physical models are coupled by interface normalization to obtain a digital twin of the VOCs equipment's full-domain multi-physics field. S1.2 Align the clocks of the physical device acquisition terminal and the twin simulation server using the IEEE 1588v2 precision clock synchronization protocol, divide the data transmission and reception time slots into fixed equal lengths, configure a bidirectional point-to-point data transmission channel, and establish a bidirectional data interaction link between the physical device and the twin. S1.3. Using the same source timestamp marking method, combined with the sliding timing alignment algorithm, the real-time operation data of physical equipment, exhaust emission monitoring data and twin fault-free standard simulation data are uniformly time-calibrated to complete the synchronization and alignment of the three types of data.

[0010] Preferably, in S2, the normal monitoring branch extracts the virtual and real dual-domain co-source operating parameters according to a unified sampling frequency, performs difference calculation to obtain the original residual, and uses an adaptive wavelet threshold denoising algorithm to remove environmental noise and non-fault interference residuals corresponding to operating condition switching, and outputs normal static residual data; when the normal static residual data does not exceed the safety threshold, the normal monitoring branch continues to operate, determines that the equipment is currently in a normal state or an acceptable slight drift state, and continuously updates the normal static residual data; The abnormal disturbance branch continuously monitors the change in normal static residual data, and introduces an exponential moving average discrimination algorithm to determine the state of normal static residual data exceeding the limit. When the normal static residual data continuously exceeds the safety threshold, the twin generates a small-amplitude micro-disturbance signal according to a preset amplitude and preset duration, and sends it to the control terminal of the physical equipment synchronously. All operating parameters of the physical equipment after being disturbed are collected as disturbance dynamic response data.

[0011] Preferably, in step S3, a principal component analysis dimensionality reduction and fusion algorithm is used to perform dimension alignment and fusion processing on the normal static residual data and the disturbance dynamic response data. The fused feature vector is then input into a pre-stored directed graph of equipment fault propagation. Based on the failure logic of nodes within the graph, the following are determined: a single component hard fault corresponding to instantaneous node failure, a multi-component coupled soft fault corresponding to multi-node linkage offset, and a process parameter drift latent fault corresponding to slow node parameter drift. Real-time detection values ​​of exhaust gas emissions are retrieved, VOCs emission concentration offset is calculated, and a pre-set grading standard is matched using a grading threshold interval matching algorithm to complete the determination of the environmental impact level of the fault.

[0012] Preferably, in S4, the virtual-real bidirectional closed-loop correction simulation is divided into two fixed execution processes: forward backtracking simulation and reverse parameter correction. The forward backtracking simulation takes the fault characteristics as input and combines the graph path reverse traversal algorithm to traverse all related nodes along the directed graph of fault propagation in reverse, locate the original cause of the fault, and sort out the fault propagation path step by step. The reverse parameter correction takes the real disturbance dynamic response data of the physical equipment as the benchmark, and uses the gradient descent adaptive calibration algorithm to fine-tune the internal coupling coefficient of the twin in groups, and batch corrects the multi-physics coupling parameters of the twin to offset the inherent modeling error of the twin model.

[0013] Preferably, in step S5, based on the two output dimensions of fault location results and fault environmental impact level, a hierarchical analysis decision algorithm is used to match and generate an emergency control plan without shutdown and a maintenance plan without shutdown, respectively; a dedicated time-series database is constructed, and disturbance dynamic response data, fault-related emission data, and model calibration data are entered in a unified data format; based on the historical data in the database, an incremental streaming learning algorithm is used to iteratively update the topology nodes of the fault propagation directed graph, and synchronously iteratively optimize the internal coupling parameters of the twin simulation model to achieve autonomous iteration of diagnostic capabilities.

[0014] Preferably, the system includes: a multiphysics twin modeling synchronization module, a dual-mode parallel diagnosis module, a fault tracing and classification module, a virtual-real closed-loop simulation correction module, and a diagnostic data iteration module; The multi-physics twin modeling synchronization module is used to construct a full-domain multi-physics digital twin, establish a two-way data interaction link, and synchronously acquire real-time operating data of physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of the twin. The dual-mode parallel diagnostic module is used to run a passive residual monitoring and active micro-disturbance dual parallel diagnostic architecture, calculate the virtual and real dual-domain operating parameter residuals, separate non-fault interference residuals, obtain normal static residual data, send out controllable small-amplitude micro-disturbance signals and collect disturbance dynamic response data. The fault tracing and classification module is used to integrate normal static residual data and disturbance dynamic response data, match the fault propagation directed graph, distinguish three types of faults, associate VOCs emission concentration offset, and determine the environmental impact level of the fault. The virtual-real closed-loop simulation correction module is used to perform virtual-real bidirectional closed-loop correction simulation, trace back the fault causes and propagation paths, and reverse correct the twin multi-physics coupling parameters. The diagnostic data iteration module is used to output fault diagnosis results and corresponding control and maintenance plans, complete the closed-loop storage of disturbance dynamic response data, fault-related emission data and model correction data, and update the fault map and twin simulation model.

[0015] Preferably, the system data flow is as follows: the multiphysics twin modeling synchronization module collects and synchronizes real-time operating data of the physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of the twin, and transmits the synchronized full-domain data to the dual-mode parallel diagnostic module; after the dual-mode parallel diagnostic module completes the calculation of virtual and real dual-domain operating parameter residuals, separation of non-fault disturbance residuals, output of normal static residual data, and acquisition of disturbance dynamic response data, it transmits the normal static residual data and disturbance dynamic response data to the fault source tracing and classification module; after the fault source tracing and classification module completes the fault type differentiation and fault environmental impact level determination, it transmits the fault-related feature data to the virtual and real closed-loop simulation correction module; after the virtual and real closed-loop simulation correction module completes the fault path backtracking and twin parameter correction, it transmits the entire diagnostic process data to the diagnostic data iteration module; the diagnostic data iteration module completes the diagnostic result output, data closed-loop storage, and fault map and twin simulation model update.

[0016] Beneficial effects This invention provides a method and system for fault diagnosis of VOCs treatment equipment based on digital twins. It has the following beneficial effects: 1. This invention employs the finite volume method, finite element transient solution algorithm, Kalman timing filter algorithm, and Arrhenius dynamics fitting algorithm to couple and construct a multi-physics field global twin model of fluid field, thermal field, electrical control signal, and catalytic reaction dynamics. Simultaneously, it combines the IEEE 1588v2 precise clock synchronization protocol and sliding timing alignment algorithm to complete global data timing calibration, achieving complete replication of multi-dimensional equipment operating conditions and high-precision synchronization of virtual and real data. This effectively solves the problems of incomplete physical field coverage and insufficient timing synchronization accuracy of virtual and real data in existing digital twin models, significantly improving the accuracy and reliability of the basic data source for fault diagnosis.

[0017] 2. This invention establishes a dual-parallel diagnostic architecture combining passive residual monitoring and active micro-perturbation. It relies on an adaptive wavelet threshold denoising algorithm to remove interference signals from the field operating conditions, and combines a fault propagation directed graph, a principal component analysis dimensionality reduction and fusion algorithm, and a hierarchical threshold interval matching algorithm to accurately distinguish three types of faults and classify the environmental risks of VOCs emissions. This effectively solves the problems of existing technologies such as single fault monitoring methods, weak anti-interference ability, coarse fault classification, and lack of environmental risk assessment. It significantly reduces the probability of false alarms and missed alarms in fault diagnosis and improves the practicality and guiding value of diagnostic results.

[0018] 3. This invention introduces a virtual-real bidirectional closed-loop correction process that combines forward fault tracing and reverse parameter correction. Combined with an incremental streaming learning algorithm, it enables online autonomous iterative updates of fault maps and twin models. This allows for continuous optimization of model parameters without manual intervention, effectively solving the problems of existing twin models lacking closed-loop correction capabilities, having fixed model parameters, and accumulating errors over long-term operation. It ensures the continuous stability of fault diagnosis accuracy throughout the entire lifecycle of the equipment and reduces the costs of manual operation and maintenance and model maintenance. Attached Figure Description

[0019] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a flowchart of the process steps of the present invention; Figure 3 This is a schematic diagram of the real-time operation data monitoring interface of the system in this invention; Figure 4 This is a schematic diagram of the VOCs emission monitoring interface of the system in this invention; Figure 5 This is a schematic diagram of the fault diagnosis result interface of the system in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1: This embodiment provides a fault diagnosis method and system for VOCs treatment equipment based on digital twins. It selects a mainstream fixed-bed catalytic oxidation VOCs treatment device as the physical entity. The device's rated design processing air volume is 10000 Nm³ / h, and the core reaction unit is a fixed-bed catalytic oxidation reactor, commonly used for end-of-pipe treatment of organic waste gas in the coating and printing industries. Figure 1 As shown, a system framework diagram is provided, illustrating the system's composition. The entire fault diagnosis system consists of a multiphysics twin modeling synchronization module, a dual-mode parallel diagnosis module, a fault tracing and grading module, a virtual-real closed-loop simulation correction module, and a diagnostic data iteration module. These five modules work collaboratively in sequence along a fixed-time link, with each module sequentially executing all method steps S1-S5, as follows: Figure 2 As shown, a flowchart is provided to illustrate the overall process of this solution. The following is a detailed explanation of each step: First, S1 is executed independently through the multiphysics twin modeling synchronization module in the system. This module is the underlying foundational support module of the entire fault diagnosis system, undertaking all the basic work of full-domain data acquisition, multiphysics modeling, virtual-real communication establishment, and multi-channel data timing alignment. The overall hardware and software architecture of the module is as follows: The module's hardware components include a multi-source sensing sensor group, an IEEE 1588v2 high-precision clock synchronizer, an industrial edge gateway, an edge GPU parallel computing card, an industrial touch screen host computer, and a cloud-based twin simulation server. The multi-source sensing sensor group is further subdivided into pipeline pressure transmitters, vortex flow sensors, layered K-type thermocouples, an online VOCs concentration analyzer, and multi-channel electronic control signal acquisition boards. All sensing hardware is uniformly deployed in the VOCs treatment equipment's inlet and outlet pipelines, the upper, middle, and lower three-layer bed of the catalytic oxidation reactor, and key process points in the equipment's main control cabinet. This provides comprehensive coverage for collecting raw data on pipeline fluid parameters, reactor temperature parameters, equipment electronic control operating parameters, and exhaust gas concentration parameters. Simultaneously, the edge gateway performs preliminary noise reduction and preprocessing of the raw data locally.

[0022] The module's supporting software includes an industrial Linux real-time operating system, multiphysics integrated simulation and solution software, an IEEE 1588 clock synchronization driver, a sliding timing alignment program, and a multi-model interface normalization coupling program. A fixed 200ms data transmission cycle is maintained between all modules, ensuring complete synchronization of timing across on-site sensing hardware, edge computing hardware, and cloud simulation servers. This avoids the time difference issues associated with data transmission from heterogeneous devices. The entire hardware and software collaborative architecture can adapt to complex on-site conditions such as diurnal fluctuations in VOCs treatment equipment exhaust gas concentrations, seasonal changes in ambient temperature, and switching of manual operating conditions.

[0023] The overall input for this step is the parameters of the as-built static structural drawings of the VOCs treatment equipment, the heterogeneous raw data of the equipment operation on site, and the standard simulation benchmark parameters of the equipment with no faults stored in the cloud. The overall output is a full-domain digital twin of four fields coupled together: fluid field, thermal field, electronic control signal field, and catalytic reaction kinetic field; a two-way point-to-point virtual-physical communication link between the physical equipment and the twin; and a three-way standardized operation dataset with complete time alignment.

[0024] Step S1 specifically involves: constructing a full-domain multiphysics digital twin of the VOCs equipment that integrates fluid field, thermal field, electronic control signals, and catalytic reaction kinetics; establishing a two-way data interaction link between the physical equipment and the twin; and simultaneously acquiring real-time operating data of the physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of the twin, as detailed below:

[0025] Step 1: Construct a full-domain multi-physics digital twin. Using an industrial touchscreen host computer to import complete as-built drawings of the equipment, all physical structural dimensions of the equipment are replicated. Edge GPU parallel computing cards provide simulation computing power support, and four independent mechanism models—fluid flow, thermal field, electrical control signals, and catalytic reaction kinetics—are built. Then, a software interface normalization program is used to unify and couple the communication protocols, simulation step sizes, and data formats of the four models, eliminating communication barriers between the heterogeneous models. This fully replicates the internal waste gas flow, bed temperature changes, electrical control start-stop actions, and the entire catalytic oxidation reaction process of the physical equipment, achieving complete consistency between the virtual twin and the physical equipment's structure and operating mechanism.

[0026] Step 2: Establish a virtual-physical bidirectional data exchange link. Relying on the IEEE 1588v2 high-precision clock synchronizer to unify the global hardware clock, independent uplink and downlink transmission time slots are allocated, and physically isolated bidirectional data transmission channels are built. The uplink channel is responsible for uploading actual field operation data from the physical device to the twin, while the downlink channel is responsible for sending simulation control commands and micro-disturbance signals from the twin to the physical device, achieving zero-delay bidirectional data exchange between the virtual and physical ends.

[0027] It should be noted that micro-disturbance refers to small, controllable disturbances under operating conditions with amplitude controlled within ±3% of the equipment's rated parameters. These disturbances will not cause equipment shutdowns or affect the compliance of exhaust gas emissions with standards. They are only used to stimulate potential fault characteristics of the equipment.

[0028] Step 3: Synchronously acquire three types of common source operational data. Real-time operational data of physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of twins are uniformly collected through industrial edge gateways. The timestamps of the three data streams are unified by a sliding time sequence alignment algorithm to compensate for transmission time sequence deviations. Finally, three types of common source datasets with time sequence alignment and uniform format are obtained, providing a standard data foundation for subsequent virtual-real residual comparison and fault monitoring.

[0029] Step S1 also includes three sub-steps: S1.1, S1.2, and S1.3. The specific contents of each sub-step are as follows: S1.1 The fluid field mesh modeling is completed using the finite volume method, the transient temperature field mapping modeling is completed using the finite element transient solution algorithm, the electronic control signal timing synchronization modeling is completed using the Kalman time-series filtering algorithm, and the catalytic reaction kinetic parameter fitting modeling is completed using the Arrhenius dynamics fitting algorithm. The four independent physical models are coupled by interface normalization to obtain a digital twin of the VOCs equipment's full-domain multi-physics field.

[0030] This process relies on an industrial touch-screen host computer to complete the drawing analysis and model initialization, while the edge GPU parallel computing card synchronously and in parallel completes the simulation calculations of four types of physical fields, without the need for manual step-by-step intervention. The detailed operating logic and calculation formulas of each modeling step are as follows: ① Fluid field modeling: The finite volume method is used to perform hexahedral structured meshing on the entire space of the waste gas pipeline and reactor cavity. Based on the fluid mass conservation equation and momentum conservation equation, the waste gas velocity and pipeline static pressure distribution are solved to restore the real fluid flow state inside the equipment. The specific calculation steps of the finite volume method are as follows: Step 1: Discretize the reactor fluid computational domain into several non-overlapping finite control volumes; Step 2: Spatial discretize the control volume fractal conservation equations and solve for the interface flux; Step 3: Complete the transient iterative calculation using a time-progression scheme; Step 4: Output the overall flow velocity and pressure distribution results after iterative convergence.

[0031] The fluid mass conservation equation and momentum conservation equation are as follows: Mass conservation integral control equations describe the conservation relationship of fluid mass within the control volume and are the basis for fluid field modeling. They are used to solve the flow distribution of exhaust gas in pipes and reactors.

[0032] The specific form of integration is as follows: ; in: Regarding time The partial derivative operator represents the instantaneous rate of change of a physical quantity over time. : Control volume Perform volume fractionation; : Density of the waste gas medium; : Control volume micro-element; : Control body surface Perform area integration; : Fluid velocity vector, representing the direction and magnitude of exhaust gas flow; : The unit vector of the outer normal to the surface of the control volume, pointing outward from the surface; : Controlling the surface area of ​​the body.

[0033] Momentum conservation integral governing equations: used to solve the momentum change during fluid motion, taking into account the effects of pressure, viscous forces and volume forces, etc., suitable for simulating complex flow fields in reactors, and providing accurate pressure and velocity field mappings for twins.

[0034] The specific form of integration is as follows: ; in: Regarding time The partial derivative operator represents the instantaneous rate of change of a physical quantity over time; : Control volume Perform volume integration; : Density of the waste gas medium; : Fluid velocity vector; : Control volume micro-element; : Control body surface Perform area integration; Convection term (momentum convection flux) represents the convective transport of momentum through a surface; The dot product of the velocity vector and the surface normal represents the velocity component perpendicular to the surface. : Surface force term generated by pressure gradient (negative sign indicates that the pressure direction is opposite to the outward normal); Hydrostatic pressure (unit: Pa); : Unit vector of the outer normal to the surface of the control volume; : Control volume surface area micro-element; Surface force term generated by viscous stress; : Viscous stress tensor (second-order tensor), which describes the viscous shear force inside a fluid.

[0035] ② Thermal field modeling: The porous media region of the catalytic bed is divided into tetrahedral unstructured meshes using a finite element transient solution algorithm. With a fixed simulation step size of 1 second, heat sources from waste gas convection and internal electric heating are superimposed. The temperature field of the entire reactor is solved time-by-time, realizing a real-time one-to-one mapping between the temperature of the physical equipment and the twin temperature. The calculation steps corresponding to the finite element transient solution algorithm are as follows: Step 1: Mesh the solid and fluid domains of the catalytic bed and construct the finite element matrix; Step 2: Assemble the global temperature stiffness matrix and heat source load vector; Step 3: Introduce a time discretization scheme and complete the recursive solution of the transient term; Step 4: Apply wall and inlet boundary conditions and iteratively solve the global transient temperature field.

[0036] The specific calculation formula is as follows: ; In the formula: The specific heat capacity of the solid in the catalytic bed; This is the partial differential term of temperature with respect to time, characterizing the instantaneous rate of change of bed temperature; The thermodynamic temperature of the grid node; The thermal conductivity of the catalytic bed; For Hamiltonian gradient operators; The power density of the heat source for waste gas convection heat transfer; The power density of the internal heat source for electric heating.

[0037] ③ Electronic control signal timing synchronization modeling: The Kalman timing filter algorithm is used to filter noise and calibrate the timing of voltage, current, and equipment switching signals acquired by the electronic control acquisition board, eliminate circuit transmission pulse interference, unify the timing of multiple electronic control signal acquisitions, and ensure complete alignment of virtual and real electronic control signal timings. The specific calculation steps of the Kalman timing filter algorithm are as follows: This algorithm consists of two core stages: prediction and update, and five closed-loop iterative operations: Step 1 (State Prediction): Based on the optimal electronic control state at the previous moment, predict the prior state value at the current moment; Step 2 (Covariance Prediction): Simultaneously predict the state covariance matrix at the current moment, representing the range of prediction error fluctuation; Step 3 (Filter Gain Calculation): Combine observation noise and process noise to solve for the optimal Kalman gain, balancing the weights of predicted and measured values; Step 4 (State Update): Use the actual measured electronic control observations on site to correct the prior state and obtain the optimal posterior state at the current moment; Step 5 (Covariance Update): Update the posterior covariance matrix and enter the next sampling cycle for iterative operation.

[0038] The specific calculation formula is as follows: State prior prediction: ; Covariance prior prediction: ; Kalman gain calculation: ; Post-state update: ; Posterior update of covariance: ; In the formula: The prior state estimate of the electronic control signal at time k; The optimal posterior state value at time k-1; This is the system state transition matrix; The state covariance matrix; The process noise covariance matrix; K is the Kalman filter gain at time k; H is the observation matrix; To measure the noise covariance matrix; denoted as k, where k represents the actual measured value of the electronic control signal at time k; I represents the identity matrix.

[0039] ④ Catalytic reaction kinetic modeling: The Arrhenius kinetic fitting algorithm is used, based on the on-site measured inlet and outlet VOCs concentrations and bed reaction temperature data, to fit the core kinetic parameters of the reaction activation energy and pre-exponential factor, and to restore the actual catalytic oxidation reaction rate variation law. The specific calculation steps of the Arrhenius kinetic fitting algorithm are as follows:

[0040] Step 1: Input the real-time bed temperature and gas constant, and substitute them into the kinetic formula to calculate the reaction rate on the simulation side; Step 2: Match the true value of the reaction rate measured in the field under the same working conditions; Step 3: Construct the mean square error loss function to quantify the deviation between the simulation and measured data; Step 4: Correct the pre-exponential factor and activation energy parameters inside the kinetic model in reverse; Step 5: Iterate until the fitting error reaches the standard, and complete the calibration of the reaction model parameters.

[0041] The specific calculation formula is as follows: ; In the formula: The rate constant for VOCs catalytic reaction; Pre-exponential factors; This is the natural exponent operation function; It is the activation energy of the reaction; This is the universal gas constant; The temperature of the reactor bed is the thermodynamic temperature. The total amount of valid sample data; For the first Group simulation reaction rate; The measured reaction rate for the i-th group; .

[0042] After completing the independent modeling of the four types of models, the software automatically completes the interface normalization coupling, unifies the model communication protocol and simulation step size, and finally generates a global integrated multiphysics digital twin.

[0043] S1.2 Align the clocks of the physical device acquisition terminal and the twin simulation server using the IEEE 1588v2 precise clock synchronization protocol, divide the data transmission and reception time slots into fixed equal lengths, configure a bidirectional point-to-point data transmission channel, and establish a bidirectional data interaction link between the physical device and the twin.

[0044] The hardware of this project relies on the IEEE 1588v2 high-precision clock synchronizer to perform global clock calibration on all sensors, edge gateways, and cloud-based twin simulation servers, controlling the clock synchronization error of all devices to within 1ms and eliminating inherent hardware clock deviations. The software relies on the IEEE 1588 clock synchronization driver to divide uplink and downlink dedicated data transmission and reception time slots with consistent durations and no interference, avoiding uplink and downlink data competing for bandwidth and causing transmission lag. Finally, a physically isolated point-to-point bidirectional transmission channel is built. The uplink channel is responsible for uploading physical device operation data, and the downlink channel is responsible for issuing twin control commands, stably realizing real-time bidirectional data interaction between the virtual and physical ends.

[0045] S1.3. Using the same source timestamp marking method, combined with the sliding timing alignment algorithm, the real-time operation data of physical equipment, exhaust emission monitoring data and twin fault-free standard simulation data are uniformly time-calibrated to complete the synchronization and alignment of the three types of data.

[0046] The hardware in this sub-step collects three heterogeneous raw data streams through an industrial edge gateway. The software adds a unified, source-same timestamp to all raw data. Then, a sliding time-series alignment algorithm is used to compensate for the time-series offsets generated during the transmission of multi-channel data, eliminating data misalignment issues caused by different acquisition hardware and simulation modules. This ensures complete time-dimensional matching of subsequent virtual-real residual comparison data. The specific calculation steps of the sliding time-series alignment algorithm are as follows: Step 1: Read the original timestamps of the physical device and the digital twin model respectively; Step 2: Calculate the time deviation between the two sets of time series data; Step 3: Perform linear compensation on the original timestamps of the device based on the timing of the twin model; Step 4: Unify the global sampling clock to complete the one-to-one alignment of the timing of the virtual and physical data.

[0047] The specific calculation formula is as follows: ; In the formula: To standardize the timestamp after compensation; This is the timestamp of the original data collection. This is the timing offset compensation amount.

[0048] Secondly, S2 is executed independently through the dual-mode parallel diagnostic module in the system. This module is the core fault monitoring unit of the system. It adopts a dual-branch parallel interference-free operation mode. The normal monitoring branch and the abnormal disturbance branch are calculated synchronously and run independently, without competing for computing power and transmission bandwidth. The overall hardware and software architecture of the module is as follows:

[0049] The dual-mode parallel diagnostic module hardware consists of an industrial edge computing gateway, a native PLC control cabinet for the equipment, an electric heating power regulator, and an electric intake flow regulating valve. The industrial edge computing gateway is deployed at the field edge and is responsible for local real-time residual calculation, noise removal, and anomaly detection, avoiding the monitoring lag problem caused by cloud data transmission delay. The PLC control cabinet serves as the field execution center, receiving various control commands and micro-disturbance signals issued by the twin. The electric heating power regulator and the electric intake flow regulating valve are field execution components used to execute small, controllable operating condition disturbances. The disturbance amplitude of all execution components is limited to ±3% of the equipment's rated parameters, which will not change the normal treatment conditions of the equipment and will not affect the emission compliance effect of the waste gas treatment.

[0050] The dual-mode parallel diagnostic module's supporting software includes a virtual and real parameter residual calculation program, an adaptive wavelet threshold denoising program, an exponential moving average discrimination program, and a micro-disturbance signal generation and distribution program. The two diagnostic branches run independently in the background, sharing the time-aligned same-source dataset output from step S1. There is no computing power contention or data interference between the branches, which is suitable for the 24 / 7 continuous operation requirements of VOCs treatment equipment.

[0051] The overall input for this step is the timing-aligned virtual-real source operating parameters output by S1; the overall output is the normal static residual data after removing environmental interference, the equipment operating status judgment result, and the physical equipment's global disturbance dynamic response data.

[0052] It should be noted that the normal monitoring branch is a passive monitoring branch, which does not require the issuance of control signals and operates under no-load for a long time; the abnormal disturbance branch is an active excitation branch, which is activated only after the limit is exceeded, so as to avoid long-term disturbances affecting the normal operation of the equipment.

[0053] Step S2 specifically involves: establishing a dual-parallel diagnostic architecture combining passive residual monitoring and active micro-disturbance. The architecture consists of a normal monitoring branch and an abnormal disturbance branch that operate in parallel. The normal monitoring branch calculates the residuals of virtual and real dual-domain operating parameters in real time, separates the residuals of environmental noise and non-fault interference caused by operating condition switching, and obtains normal static residual data. When the normal static residual data does not exceed the safety threshold, the normal monitoring branch continues to operate, determines that the equipment is currently in a normal state or an acceptable slight drift state, and continuously updates the normal static residual data. When the normal static residual data exceeds the safety threshold, the abnormal disturbance branch is triggered, and a controllable small-amplitude micro-disturbance signal is sent by the twin to synchronously act on the physical equipment to obtain the dynamic response data of the disturbance.

[0054] The routine monitoring branch extracts operating parameters from both the real and virtual domains simultaneously at a pre-set unified sampling frequency. It performs a difference calculation by mapping the actual measured operating parameters of the physical equipment to the fault-free standard simulation parameters of the twin, obtaining an original residual sequence containing effective fault residuals, environmental noise residuals, and operating condition switching interference residuals. Subsequently, an adaptive wavelet threshold denoising algorithm is called to accurately remove non-fault interference residuals corresponding to two scenarios: environmental temperature fluctuations and on-site artificial waste gas treatment load switching. Invalid noise components are removed, retaining only the effective residual signals that reflect abnormal equipment operation, ultimately outputting clean and interference-free routine static residual data. The system compares the routine static residual data with a locally preset fault safety threshold in real time. When the routine static residual data does not exceed the safety threshold, the system continues to maintain closed-loop operation of the routine monitoring branch, determining whether the equipment is currently in normal operation or in an acceptable state of slight parameter drift. Simultaneously, it continuously iterates and updates the routine static residual baseline data to adapt to the slow baseline drift caused by long-term equipment operation, avoiding baseline aging that could lead to subsequent fault misjudgments. The safety threshold is divided into instantaneous threshold and continuous threshold. Exceeding the limit at a single sampling point will not be considered abnormal, thus avoiding instantaneous pulse interference.

[0055] Status determination logic: If the normal static residual does not exceed the preset safety threshold, the device is determined to have normal / slightly acceptable parameter drift, and the residual baseline is continuously updated; if the normal static residual exceeds the safety threshold, a persistent anomaly is determined, and preparations are made to trigger the disturbance branch. The specific calculation steps of the adaptive wavelet threshold denoising algorithm are as follows: Step 1: Perform multi-scale wavelet decomposition on the original noisy residual signal to separate the low-frequency effective signal and the high-frequency noise component; Step 2: Calculate the standard deviation of high-frequency noise through unbiased median estimation; Step 3: Adaptively calculate the dynamic denoising threshold in combination with the signal sampling length; Step 4: Perform soft threshold shrinkage on the wavelet high-frequency coefficients according to the threshold; Step 5: Reconstruct the wavelet signal and output the clean residual signal after removing interference.

[0056] It should be noted that the virtual and real dual domains refer to two operating spaces: the physical equipment (the actual VOCs treatment equipment on site) and the digital twin virtual model (the same equipment model simulated and replicated in the cloud).

[0057] The residual refers to the difference between the simulated operating parameters of the virtual model and the measured operating parameters of the physical equipment. The larger the difference, the greater the deviation between the virtual and real operating states.

[0058] The abnormal disturbance branch continuously monitors the real-time changes of the normal static residual data without interruption, tracking two key indicators: residual fluctuation amplitude and fluctuation duration. Simultaneously, an exponential moving average discrimination algorithm is introduced to smooth instantaneous pulse fluctuations in the residuals, filter out millisecond-level instantaneous abnormal spikes, accurately determine the continuous exceeding state of the normal static residual data, and distinguish between instantaneous pseudo-anomalies and real, continuous fault anomalies. When the algorithm determines that the normal static residual data continuously exceeds the preset safety threshold and eliminates instantaneous interference, a hard-linkage trigger is immediately activated in the abnormal disturbance branch. The cloud-based twin simulation server generates a small-amplitude micro-disturbance signal with controllable amplitude, fixed duration, and no secondary interference, based on the pre-defined disturbance amplitude and preset disturbance duration. This signal is synchronously sent to the PLC control terminal of the physical equipment and the field execution components via a virtual-physical bidirectional point-to-point communication link. Simultaneously, all operating parameters of the physical equipment after the disturbance are completely collected, and this complete set of parameters is used as the dynamic response data for the disturbance, completing the data collection and storage of the entire abnormal branch. The calculation steps of the exponential moving average discrimination algorithm are as follows: Step 1: Preset fixed smoothing weights and allocate weight ratios between real-time residuals and historical residuals; Step 2: Retrieve the original residuals at the current time and the historical smoothed residuals at the previous time; Step 3: Weightedly fuse the two sets of residual data and filter out instantaneous impulse interference; Step 4: Output a continuously smoothed residual curve and determine whether the residuals continuously exceed the limits.

[0059] The specific calculation formula is as follows: ; In the formula: This represents the static residual value after smoothing at the current moment. The smoothing weight coefficient is updated in real time for the residuals, with a fixed value of 0.2, which takes into account both historical and real-time residual weights; This represents the original, unsmoothed, normal static residual at the current moment. This represents the historical smoothed residual value from the previous sampling time. For historical residual backtracking weights.

[0060] Next, S3 is executed independently through the fault tracing and classification module in the system. This module receives two core fault data streams from S2: normal static residual data and disturbance dynamic response data. It completes the closed-loop processing of fault data fusion, fault type identification, and environmental risk classification. The overall hardware and software architecture of the module is as follows:

[0061] The module's hardware components include a cloud-based parallel computing server and an industrial solid-state storage hard drive. The industrial solid-state storage hard drive locally pre-stores a complete directed graph of equipment fault propagation and a three-level VOCs environmental emission risk threshold comparison table, enabling rapid local matching of fault characteristics without needing to retrieve the graph from the cloud, thus reducing response latency. The cloud-based parallel computing server is responsible for complex matrix operations such as dimensionality reduction and feature fusion of high-dimensional fault data, ensuring real-time fault identification and meeting the needs of on-site online diagnosis.

[0062] The module's supporting software includes a principal component analysis dimensionality reduction and fusion program, a fault map topology matching program, an emission offset calculation program, and an automatic matching program for classification thresholds. The software can automatically complete the dimensionless processing of high-dimensional data, fault topology node matching, and automatic classification of emission risks without the need for manual intervention, making it suitable for unattended operation and maintenance scenarios.

[0063] The overall input for this step is the normal static residual data, disturbance dynamic response data, and real-time VOCs exhaust emission monitoring data output by S2; the overall output is the precise fault type, fault location, and the corresponding VOCs environmental impact risk level.

[0064] Step S3 specifically involves: combining normal static residual data with dynamic response data to match the directed fault propagation graph of the equipment, distinguishing between single-component hard faults, multi-component coupled soft faults, and latent faults caused by process parameter drift, and simultaneously correlating VOCs emission concentration offsets to determine the environmental impact level of the fault. It should be noted that the directed fault propagation graph refers to a pre-built equipment fault topology network diagram with propagation directions, where nodes correspond to equipment monitoring parameters / hardware components, and lines represent fault propagation paths.

[0065] The following is the specific execution process: Step 1: Dimensional Alignment and Fusion of Heterogeneous Fault Data from Two Paths. A principal component analysis (PCA) dimensionality reduction and fusion algorithm is employed to perform standardization preprocessing and dimension alignment / dimensionality reduction fusion on both the normal static residual data and the dynamic response data under disturbance, which have inconsistent dimensions and different units of measurement. This eliminates the problems of unit of measurement conflicts and dimension mismatches between the two heterogeneous data streams, removes redundant feature information within both data streams, compresses the volume of invalid data, and eliminates the identification errors caused by direct matching of heterogeneous data. Finally, a standardized fault feature vector with unified dimensions, consistent units of measurement, and simplified features is generated.

[0066] The specific calculation steps of the principal component analysis dimensionality reduction and fusion algorithm are as follows: Step 1: Standardize and normalize the residual feature data of the two channels with different dimensions; Step 2: Calculate the global feature covariance matrix to characterize the correlation degree of each parameter; Step 3: Perform eigenvalue decomposition on the covariance matrix; Step 4: Filter the principal components with high contribution rate and remove redundant noise features; Step 5: Complete the dimensionality reduction of high-dimensional data and output a unified dimension fault feature vector.

[0067] The specific calculation formula is as follows: ; In the formula: This refers to standardized dimensionless fault characteristic data. ; This is a vector of the mean values ​​of a single set of fault data. The standard deviation of a single set of fault data; The covariance matrix of the fault data represents the correlation of fault features in different dimensions. This represents the total number of fault data samples participating in the fusion process in a single batch. This represents the original fault sample data for the i-th group; This is the matrix transpose operator.

[0068] Step 2: Fault Map Import and Precise Differentiation of Three Fault Types. The fused standardized fault feature vectors are fully input into the pre-stored directed fault propagation graph on the local industrial solid-state drive. Based on the graph's pre-defined node failure logic and fault propagation link rules, the three types of faults are accurately classified without omission: A sudden, instantaneous failure of a single monitoring node in the graph corresponds to a single-component hard fault (sudden damage to a single component); synchronized offsets and coupled fault signals across multiple related nodes in the graph correspond to multi-component coupled soft faults caused by mutual interference; and no sudden, instantaneous changes in nodes, only slow, long-term parameter drift, correspond to latent process parameter drift faults where there is no hardware damage but only deviations in operating process parameters. The following provides a detailed explanation of the three types of faults: 1. Single component hard failure: ① Graph node characteristics: A single independent monitoring node experiences a millisecond-level step-like instantaneous change, without a gradual change in preceding parameters, and all surrounding related nodes show no linkage fluctuations; ② Fault mechanism: The equipment hardware itself suffers sudden hardware damage such as irreversible physical breakdown, open circuit, or burnout; ③ On-site equipment manifestations: Sensor malfunction, heating module open circuit, solenoid valve jamming, relay breakdown, and other obvious hardware failures, causing immediate partial operational failure of the equipment; ④ Diagnostic difficulties: The fault characteristics are obvious, but the source of the fault cannot be located; ⑤ Environmental impact: VOCs emission concentration rises rapidly in a short period of time, with an extremely high risk of exceeding environmental standards; ⑥ Typical case: The temperature sensor suddenly breaks the circuit, and the temperature monitoring node instantly returns to zero.

[0069] 2. Multi-component coupled soft faults: ① Graph node characteristics: No single node experiences independent abrupt changes; multiple upstream and downstream related nodes synchronously and slightly shift in the same direction; fault signals between nodes are coupled and superimposed, and the fault exhibits a linkage transmission pattern; ② Fault mechanism: All hardware components are not physically damaged, but the operating parameters of multiple components interfere with each other, and the coordinated operating conditions are unbalanced, which belongs to the system-level operational coupling anomaly, rather than hardware damage; ③ Field equipment manifestations: Inlet air flow, bed temperature, and heating power shift synchronously and slowly; individual component detection shows no abnormalities, but deviations occur in the coordinated operation of multiple components; ④ Diagnostic challenges: No explicit fault alarms; traditional single-parameter monitoring cannot identify coupled correlation anomalies; ⑤ Environmental impact: Waste gas treatment efficiency continuously declines, and VOCs emissions rise slowly and uniformly; ⑥ Typical case: Fluctuations in inlet air flow are linked to fluctuations in catalytic bed temperature, resulting in dual-parameter coupled shifts.

[0070] 3. Latent Faults Due to Process Parameter Drift: ① Characteristics of Spectrum Nodes: No instantaneous changes or node-linked fluctuations occur at any monitoring node. All node parameters drift continuously, unidirectionally, and slowly with a small amplitude over the duration of operation. ② Fault Mechanism: Long-term operation of the equipment leads to chronic aging, including catalyst activity decay, reduced heat dissipation due to dust accumulation in pipelines, and a slow increase in heat loss in the equipment cavity. No hardware damage occurs throughout the process. ③ On-site Equipment Performance: No alarm pop-ups are displayed on the equipment, and operating parameters do not fluctuate drastically. Maintenance personnel cannot detect abnormalities through visual inspection or conventional monitoring. ④ Diagnostic Difficulty: The most concealed type of fault, completely undetectable by conventional threshold monitoring. ⑤ Environmental Impact: Long-term small deviations in VOC emissions are difficult to detect, and long-term accumulation can cause environmental hazards. ⑥ Typical Case: After 30 days of continuous operation, catalyst activity slowly decreases, and reaction parameters continue to drift.

[0071] Step 3: VOCs emission data retrieval and environmental risk level determination. The system retrieves real-time emission data from the online VOCs concentration analyzer and calculates the concentration offset between the actual VOCs emission concentration under fault conditions and the baseline emission concentration under fault-free standard conditions. Then, using a tiered threshold interval matching algorithm, the calculated absolute and relative concentration offsets are matched against the system's preset Level 1, Level 2, and Level 3 environmental grading standards, automatically classifying the environmental impact level of the fault and outputting the corresponding risk level results. Environmental grading is divided into Level 1 (minor risk), Level 2 (moderate risk), and Level 3 (severe risk), corresponding to three control requirements: no shutdown required, shutdown as appropriate, and emergency shutdown, respectively. The specific calculation steps for emission concentration offset are as follows: Step 1: Collect real-time VOCs emission concentration at the site; Step 2: Retrieve the standard baseline emission concentration of the equipment under the same operating conditions; Step 3: Perform difference calculation to obtain the absolute concentration offset; Step 4: Normalize and calculate the relative offset percentage; Step 5: Compare with the classification threshold to determine the environmental risk level.

[0072] The specific calculation formula is as follows: ; In the formula: This represents the absolute offset of VOCs concentration, visually reflecting the exceedance value. This refers to the real-time VOCs emission concentration at the site under fault conditions. The standard VOCs emission concentration is the benchmark for equipment failure-free operation. The relative offset rate of VOCs concentration represents the relative severity of emissions exceeding standards.

[0073] The specific calculation steps of the hierarchical threshold interval matching algorithm are as follows: Step 1: Pre-store three fixed concentration offset threshold ranges: Level 1: Slight Risk (0, 5%), Level 2: Moderate Risk (5%, 15%), Level 3: Severe Risk (15%, +∞); Step 2: Use the concentration relative offset rate calculated above as the input feature value; Step 3: Compare the inclusion relationship between the feature value and each threshold range; Step 4: Match the corresponding environmental risk level one-to-one and simultaneously bind the corresponding operation and maintenance disposal strategy; Step 5: Lock the risk level result and transmit it to the backend.

[0074] Subsequently, S4 is executed independently through the virtual-real closed-loop simulation correction module in the system. This module is specifically designed to compensate for the inherent modeling errors caused by mesh approximation, simplified boundary conditions, and dynamic parameter fitting deviations during the long-term operation of the digital twin model, ensuring the long-term stability of the virtual-real mapping accuracy. The overall hardware and software architecture of the module is as follows:

[0075] The module's hardware components include a cloud-based parallel simulation computing unit and an independent simulation data storage module. The cloud-based parallel simulation computing unit supports high-precision simulation calculations for fault path reverse traversal and gradient iteration, meeting the computing power requirements for large-scale graph topology traversal and multi-parameter iterative calibration. The independent simulation data storage module partitions and isolates virtual and real disturbance response data, original uncorrected model parameters, and corrected model parameters to avoid cross-interference between different types of simulation data and ensure data purity throughout the calibration process.

[0076] The module's supporting software includes a fault path reverse traversal program, a loss function construction program, and a gradient descent adaptive parameter calibration program. The software is strictly divided into two fixed execution flows in sequence: forward backtracking simulation and reverse parameter correction. The flow cannot be changed. The entire process uses the actual operating data of the physical equipment as the sole truth benchmark to complete the autonomous closed-loop correction of the model.

[0077] The overall input for this step is the standardized fault feature vector output by S3 and the comparison data of disturbance response at both the virtual and real ends; the overall output is the complete temporal fault propagation path, the original root cause of the fault, and the parameters of the multiphysics coupled twin model after correction.

[0078] Step S4 specifically involves: conducting a virtual-real two-way closed-loop correction simulation based on the twin model to eliminate the inherent modeling errors of the twin model.

[0079] The specific execution process is as follows: (1) Forward backtracking simulation process: The forward backtracking simulation uses the standardized fault feature vector output by S3 as the only input data. It calls the graph path reverse traversal algorithm, starting from the external manifestation node of the fault (excessive exhaust gas concentration, abnormal bed temperature, abnormal pipeline pressure, etc.), and traverses all related topological nodes in the graph in reverse along the fault propagation directed graph. It sorts out the fault propagation path from top to bottom, distinguishes the initial original fault node of the equipment from the secondary fault nodes triggered by subsequent chains, and finally accurately locates the original root cause of the fault, clarifying the complete propagation logic of the fault from its source to its external manifestation.

[0080] The specific calculation steps of the graph path reverse traversal algorithm are as follows: Step 1: Locate the final visual anomaly monitoring node of the fault; Step 2: Based on the directed propagation edge of the graph, trace back the upstream related nodes layer by layer in reverse; Step 3: Continue to search backwards until the root node of the fault without any preceding cause; Step 4: Arrange all nodes in reverse chronological order to generate a complete set of fault propagation paths.

[0081] The specific calculation formula is as follows: ; In the formula: For a complete set of time-sequential fault back propagation paths; This serves as the terminal observation node for externalizing faults, corresponding to visible fault phenomena such as excessive exhaust gas and abnormal bed temperature. , It serves as an intermediate transitional node in the fault propagation chain; This is the original root node of the fault, i.e., the location where the fault first occurred; This is a symbol for a set of nodes.

[0082] (2) Inverse parameter correction process: The reverse parameter correction uses the actual measured data of the dynamic response of the physical equipment under real disturbances as the absolute truth benchmark, constructs the error loss function between the twin simulation data and the actual measured data, and quantifies the inherent simulation bias of the model. Then, the gradient descent adaptive calibration algorithm is used to fine-tune the coupling coefficients of the four types of models, namely the internal fluid field, thermal field, electrical control signal field and catalytic reaction kinetic field, in groups. The multi-physics field coupling simulation parameters of the twin are uniformly corrected in batches, and the virtual-real response bias is continuously reduced iteratively until the overall simulation error of the model converges to the system's preset qualified threshold. This offsets all the inherent modeling errors caused by the twin model meshing, boundary condition simplification and parameter fitting from the bottom layer of the model.

[0083] The specific calculation steps of the gradient descent adaptive calibration algorithm are as follows: Step 1: Import twin simulation data and actual measured data from the equipment; Step 2: Construct the mean squared error loss function to quantify the overall deviation between the virtual and real models; Step 3: Solve for the gradient of the loss function to determine the optimal correction direction for the parameters; Step 4: Iteratively update the model coupling parameters by combining the learning step size; Step 5: Iterate repeatedly until the error converges, then stop model correction.

[0084] The specific calculation formula is as follows: , ; In the formula: The mean square loss function for model simulation error is used to quantify the virtual-real response deviation of the twin model; This is the set of all multiphysics coupling parameters for the twin. This represents the total number of disturbance response data samples. The simulation dynamic response data for the i-th twin; This represents the true value data of the measured dynamic response of the i-th physical device. These are the new coupling parameters after iterative updates; These are the original coupling parameters before iteration; The step size for gradient iteration learning controls the magnitude of parameter correction; This is the gradient operator for the loss function, pointing in the direction where the model error decreases the fastest.

[0085] Finally, S5 is executed independently through the diagnostic data iteration module in the system. This module serves as the output terminal and self-optimization closed-loop unit of the entire diagnostic system. It receives all upstream fault diagnosis and model calibration data, completes operation and maintenance plan matching, data archiving, and system self-iterative upgrades. The overall hardware and software architecture of the module is as follows:

[0086] The module's hardware components include an industrial touch screen HMI display and an industrial-grade time-series database hard drive. The industrial touch screen HMI display is designed for on-site maintenance personnel, providing an intuitive and visual representation of the fault location, fault type, environmental risk level, and two types of maintenance solutions. The industrial-grade time-series database hard drive is used for time-series ordered storage of raw diagnostic data throughout the entire process, supporting historical fault case review, fault pattern analysis, and model iterative training.

[0087] The module's supporting software includes a hierarchical analysis decision-making program, a time-series data automatic data entry program, and an incremental streaming learning iteration program. The software runs fully automatically without human intervention, sequentially completing intelligent decision-making for operation and maintenance plans, automatic entry of full time-series data into the database, and incremental updates of fault maps and twin models, enabling the system to achieve long-term unattended autonomous optimization.

[0088] The overall input for this step is the fault tracing results, fault environmental risk level, and original time-series data of the entire process diagnosis output by S4; the overall output is a visualized fault diagnosis report, non-stop emergency control plan, shutdown maintenance plan, and iteratively optimized fault propagation directed graph and twin simulation model.

[0089] The S5 steps are as follows: output accurate fault location results, fault environmental risk level, non-stop emergency control plan and shutdown maintenance plan, and simultaneously put disturbance dynamic response data, fault-related emission data and model calibration data into the database in a closed loop, and automatically update the fault map and twin simulation model.

[0090] The specific execution process is as follows: Step 1: Dual-Dimensional Intelligent Matching of Operation and Maintenance Solutions. Strictly adhering to two fixed output dimensions—fault location results and the environmental impact level of the fault—the hierarchical analysis decision-making algorithm is invoked to comprehensively consider three evaluation indicators: the degree of equipment hardware damage, the risk of VOCs exceeding environmental emission standards, and the economic cost of on-site downtime operation and maintenance. A judgment matrix is ​​constructed to complete weight assignment and solution scoring, generating targeted matching and generation of non-stop emergency control solutions and shutdown maintenance solutions: For process parameter drift-related latent faults with no hardware structural damage and low environmental risk levels, a non-stop emergency control solution with adaptive parameter adjustment is output; for component hardware faults and faults with medium to high environmental risk levels, a precise shutdown maintenance solution including fault location, maintenance procedures, and maintenance duration is output. The specific calculation steps of the hierarchical analysis decision algorithm are as follows: Step 1: Establish a multi-dimensional evaluation index judgment matrix; Step 2: Solve for the matrix eigenvalues ​​and corresponding weight vectors to complete the matrix consistency check; Step 3: Score each of the different operation and maintenance solutions; Step 4: Combine the weighted sum of the index weights to obtain the comprehensive score of the solution; Step 5: Compare the scoring results and output the optimal operation and maintenance solution.

[0091] The specific calculation formula is as follows: , ; In the formula: A multi-indicator judgment matrix for operation and maintenance decisions, including three indicators: hardware damage due to failure, risk of exceeding environmental protection standards, and operation and maintenance costs; This is the weight vector corresponding to each evaluation indicator; To determine the largest eigenvalue of a matrix, used to verify matrix consistency; A comprehensive evaluation score is given to the operation and maintenance solution. The weight corresponding to the j-th evaluation indicator; This represents the individual score for the j-th indicator.

[0092] Step 2: Closed-loop time-series data entry for full-domain diagnostics. The system automatically collects three types of core data throughout the entire process: dynamic response data to disturbances, VOCs emission data associated with faults, and twin model calibration parameter data. It then calls the automatic time-series data entry program, relying on the module's built-in dedicated time-series database, to complete batch automatic closed-loop data entry according to a unified standardized data format and unified timestamp labels. This establishes a complete time-series fault case archive, retaining original evidence for complete fault tracing and model calibration, providing real and effective field fault samples for subsequent system iterations. The time-series database stores data in an orderly manner according to timestamps, supporting historical fault case backtracking and fault pattern analysis. Unlike ordinary relational databases, it is more suitable for continuous time-series equipment operation data.

[0093] Step 3: Autonomous Iterative Optimization of Fault Graph and Twin Model. Based on all archived historical fault data in the time-series database, an incremental streaming learning algorithm is used. Without needing to access all historical data, it incrementally updates the internal topological nodes, node association weights, and fault propagation links of the directed fault propagation graph using newly added fault samples. Simultaneously, iteratively optimizes all multi-physics coupling parameters within the twin simulation model, continuously filling in blind spots in the graph and optimizing underlying simulation parameters based on real-world fault cases. Ultimately, this achieves autonomous iterative upgrades of the entire fault diagnosis system's diagnostic capabilities.

[0094] The specific computational steps of the incremental streaming learning algorithm are as follows: Step 1: Extract newly added fault samples from the database separately, without calling all historical data; Step 2: Calculate the incremental deviation of model parameters caused by the new samples; Step 3: Add increments to the original model and graph parameters to complete fine-tuning; Step 4: Complete online incremental iteration and wait for the next batch of new data for continuous optimization.

[0095] The specific calculation formula is as follows: ; In the formula: The coupling parameters of the twin model and the node weights of the fault map are optimized through iterative optimization; These are the original model parameters and the original graph topological weights before iteration; Fine-tune the incremental values ​​of the parameters corresponding to a single round of incremental learning; For newly added samples of fault characteristics at the warehouse, iteration is based solely on the newly added data, without the need to backtrack all historical data.

[0096] like Figure 3 As shown, this is a schematic diagram of the system's real-time operating data interface. This is the front-end interface, which can monitor the operating status of physical equipment.

[0097] like Figure 4As shown, a schematic diagram of the VOCs emission monitoring interface of the system is provided to demonstrate the system's real-time monitoring of exhaust gas emission concentration.

[0098] The entire system operates according to a fixed unidirectional serial link S1→S2→S3→S4→S5, with data transmitted sequentially between modules without reverse data interference: After the multiphysics twin modeling synchronization module completes modeling and data synchronization, it transmits the standardized homogeneous dataset to the dual-mode parallel diagnostic module; the dual-mode parallel diagnostic module outputs two channels of fault residual data and then transmits them to the fault tracing and classification module; the fault tracing and classification module completes fault identification and risk classification and then transmits the data to the virtual-real closed-loop simulation correction module; the virtual-real closed-loop simulation correction module completes fault backtracking and model correction and then transmits the entire process data to the diagnostic data iteration module; finally, it completes the output of diagnostic results, data archiving, and system autonomous iteration, forming a complete closed-loop fault diagnosis process.

[0099] like Figure 5 As shown, a schematic diagram of the system's fault diagnosis results interface is provided, which is used to display the current fault status of the device on the front-end interface. Specific Implementation Example 2: This embodiment provides a practical application case of the VOCs treatment equipment fault diagnosis method and system based on digital twins in an actual industrial field. Application of fault diagnosis for RTO equipment in an automotive painting production line: A certain automotive parts painting production line uses a regenerative thermal oxidizer as the core equipment for VOCs end-of-pipe treatment, designed to handle an air volume of 15,000 Nm³ / h, mainly treating benzene series and ester organic waste gas generated in the spraying process. The equipment operates continuously for extended periods, and is susceptible to coupled, latent faults involving multiple components due to high-temperature cycling, heat storage ash accumulation, and frequent valve operations. Traditional threshold alarm systems cannot accurately distinguish the causes of these faults, leading to unintended shutdowns.

[0101] S1 step execution process: The multiphysics twin modeling synchronization module first uses the finite volume method to complete the fluid field mesh modeling, accurately reproducing the gas velocity and pressure distribution inside the furnace. It then uses a finite element transient solution algorithm to complete the thermal field mapping modeling, replicating the entire process of heat storage and release in the regenerator. Next, it uses a Kalman timing filter algorithm to complete the synchronous modeling of the electrical control signals, filtering out timing noise from the electrical control module and ensuring precise alignment of valve switching signals. Finally, it uses an Arrhenius dynamics fitting algorithm to fit the kinetic parameters of the oxidation reaction in the combustion chamber, and constructs a full-domain multiphysics digital twin through interface normalization coupling, completely replicating the equipment's full-field coupled operation characteristics. A bidirectional data interaction link is established between the physical equipment and the twin through the IEEE 1588v2 precise clock synchronization protocol and an industrial edge gateway, synchronously acquiring real-time operating data of the physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of the twin. Combined with a sliding timing alignment algorithm, high-precision timing alignment of these three types of data is achieved, eliminating diagnostic errors caused by timing deviations.

[0102] S2 step execution process: The dual-mode parallel diagnostic module establishes a dual-parallel diagnostic architecture combining passive residual monitoring and active micro-disturbance, with the two branches operating independently without interference. The normal monitoring branch calculates the residuals of operating parameters in both virtual and real domains in real time. It separates the non-fault interference residuals caused by environmental temperature and humidity fluctuations and production line load switching using an adaptive wavelet threshold denoising algorithm, and obtains the normal static residual data after removing invalid noise. Subsequently, an exponential moving average discrimination algorithm is used to determine that the residuals are in a state of continuous over-limit, rather than instantaneous operating condition fluctuation interference. Then, the abnormal disturbance branch is triggered, and the twin sends out controllable small-amplitude micro-disturbance signals (furnace switching cycle micro-disturbance, air intake flow micro-disturbance) with amplitude controlled within ±3%, while simultaneously collecting the dynamic response data of the physical equipment to uncover the coupled fault characteristics hidden behind the parameters.

[0103] S3 step execution process: The fault tracing and grading module employs principal component analysis (PCA) dimensionality reduction and fusion algorithms to align and fuse normal static residual data with dynamic response data under disturbances, eliminating the dimensional differences between the two types of data. The fused unified fault feature vector is input into the equipment fault propagation directed graph. Based on the graph node linkage offset rules, it is accurately determined to be a multi-component coupled soft fault. The specific fault mechanism is that localized ash accumulation in the regenerator leads to uneven flow distribution within the furnace, further triggering a synchronous shift in the combustion chamber temperature field; no single hardware component is damaged. The synchronously calculated VOCs emission concentration shift is 8.7%, and based on the grading threshold interval matching algorithm standard, it is determined to be a Level II moderate environmental impact.

[0104] S4 step execution process: The virtual-real closed-loop simulation correction module relies on a twin to conduct virtual-real bidirectional closed-loop correction simulations. The forward backtracking simulation calls a graph path reverse traversal algorithm, traversing all associated nodes along the directed fault propagation graph to accurately pinpoint the primary cause of the fault as ash accumulation on the upper layer of the heat storage body leading to a decrease in heat storage efficiency. Simultaneously, it comprehensively analyzes the step-by-step fault propagation path from abnormal flow rate to temperature deviation to decreased treatment efficiency. The reverse parameter correction employs a gradient descent adaptive calibration algorithm, fine-tuning the coupling parameters of the twin's fluid and thermal fields group by group based on real disturbance feedback data from the physical equipment, eliminating inherent modeling errors caused by the initial mesh generation and boundary condition simplification of the twin model.

[0105] S5 step execution process: The diagnostic data iteration module takes the fault location results and the environmental impact level of the fault obtained from this diagnosis as two core dimensions, and initiates the hierarchical analysis decision-making algorithm to conduct quantitative decision-making. The specific execution steps are as follows: Step 1: Construct a judgment matrix and set three evaluation indicators: the degree of equipment hardware damage, the risk of VOCs exceeding environmental emission standards, and the economic cost of on-site downtime operation and maintenance; at the same time, set two alternative operation and maintenance plans: Plan A is an emergency control plan without downtime and Plan B is a downtime maintenance plan.

[0106] Step 2: Assigning index weights and verifying consistency. Based on the industrial site operation and maintenance specifications, assign a weight of 0.45 to the risk of VOCs exceeding environmental emission standards, a weight of 0.35 to the degree of equipment hardware damage, and a weight of 0.20 to the economic cost of on-site downtime operation and maintenance. Complete the solution of the eigenvalues ​​of the judgment matrix and verify consistency.

[0107] Step 3: Two-dimensional item-by-item scoring: Based on the fault location results (multi-component coupled soft fault, no irreversible hardware damage): Solution A (non-stop operation) scores 85 points, Solution B scores 65 points; Based on the environmental risk level (Level 2, medium): Solution A scores 78 points, Solution B scores 82 points; Based on the operation and maintenance economic cost: Solution A scores 92 points, Solution B scores 55 points.

[0108] Step 4: Weighted comprehensive scoring. The comprehensive score of the non-stop emergency control plan is calculated to be 81.35 points, and the comprehensive score of the shutdown maintenance plan is 67.45 points. Therefore, the non-stop emergency control plan (optimizing the furnace switching cycle and strengthening the online purging of the heat storage medium) is output first, and the alternative shutdown maintenance plan (targeted disassembly and cleaning of the heat storage medium during the production downtime) is generated at the same time.

[0109] After the decision is made, the system stores the disturbance dynamic response data, fault-related emission data, and model calibration data in a closed loop into the industrial time series database. Then, using an incremental streaming learning algorithm, the system autonomously updates the fault propagation directed graph link weights and the coupling parameters of the twin simulation model based solely on the newly added fault samples, thereby achieving online autonomous iterative upgrades of the diagnostic model.

[0110] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A fault diagnosis method for VOCs treatment equipment based on digital twins, characterized in that, Includes the following steps: S1. Construct a full-domain multi-physics digital twin of VOCs equipment that integrates fluid field, thermal field, electronic control signal and catalytic reaction dynamics, establish a two-way data interaction link between the physical equipment and the twin, and simultaneously acquire real-time operating data of the physical equipment, exhaust emission monitoring data and fault-free standard simulation data of the twin. S2. Construct a dual-parallel diagnostic architecture combining passive residual monitoring and active micro-disturbance. The architecture consists of a normal monitoring branch and an abnormal disturbance branch that operate in parallel. The normal monitoring branch calculates the residuals of the virtual and real operating parameters in real time, separates the residuals of environmental noise and non-fault interference caused by operating condition switching, and obtains normal static residual data. When the normal static residual data does not exceed the safety threshold, the normal monitoring branch continues to operate, determines that the equipment is currently in a normal state or an acceptable slight drift state, and continuously updates the normal static residual data. When the normal static residual data exceeds the safety threshold, the abnormal disturbance branch is triggered, and a controllable small-amplitude micro-disturbance signal is sent by the twin to synchronously act on the physical equipment to obtain the dynamic response data of the disturbance. S3. By combining normal static residual data and disturbance dynamic response data, the directed spectrum of equipment fault propagation is matched to distinguish between single component hard faults, multi-component coupled soft faults and process parameter drift hidden faults. The environmental impact level of the fault is determined by synchronously associating the VOCs emission concentration offset. S4. Conduct virtual-real two-way closed-loop correction simulation based on twins to eliminate the inherent modeling errors of twin models; S5 outputs accurate fault location results, fault environmental risk level, non-stop emergency control plan and shutdown maintenance plan. At the same time, it puts disturbance dynamic response data, fault-related emission data and model correction data into the database in a closed loop and updates the fault map and twin simulation model autonomously.

2. The fault diagnosis method for VOCs treatment equipment based on digital twins according to claim 1, characterized in that, The specific steps of S1 are as follows: S1.1 The fluid field mesh modeling is completed using the finite volume method, the thermal transient temperature field mapping modeling is completed using the finite element transient solution algorithm, the electronic control signal time synchronization modeling is completed using the Kalman time-series filtering algorithm, and the catalytic reaction kinetic parameter fitting modeling is completed using the Arrhenius dynamics fitting algorithm. The four independent physical models are coupled by interface normalization to obtain a digital twin of the VOCs equipment's full-domain multi-physics field. S1.2 Align the clocks of the physical device acquisition terminal and the twin simulation server using the IEEE 1588v2 precision clock synchronization protocol, divide the data transmission and reception time slots into fixed equal lengths, configure a bidirectional point-to-point data transmission channel, and establish a bidirectional data interaction link between the physical device and the twin. S1.

3. Using the same source timestamp marking method, combined with the sliding timing alignment algorithm, the real-time operation data of physical equipment, exhaust emission monitoring data and twin fault-free standard simulation data are uniformly time-calibrated to complete the synchronization and alignment of the three types of data.

3. The fault diagnosis method for VOCs treatment equipment based on digital twins according to claim 1, characterized in that, In S2, the normal monitoring branch extracts virtual and real dual-domain operating parameters at a unified sampling frequency, performs difference calculations to obtain the original residual, and uses an adaptive wavelet threshold denoising algorithm to remove environmental noise and non-fault interference residuals corresponding to operating condition switching, outputting normal static residual data; when the normal static residual data does not exceed the safety threshold, the normal monitoring branch continues to operate, determines that the equipment is currently in a normal state or an acceptable slight drift state, and continuously updates the normal static residual data; The abnormal disturbance branch continuously monitors the changes in the normal static residual data and introduces an exponential moving average discrimination algorithm to determine the state of the normal static residual data exceeding the limit.

4. The fault diagnosis method for VOCs treatment equipment based on digital twins according to claim 1, characterized in that, In step S3, a principal component analysis dimensionality reduction and fusion algorithm is used to perform dimension alignment and fusion processing on the normal static residual data and the dynamic response data of the disturbance. The fused feature vector is then input into the pre-stored directed graph of equipment fault propagation. Based on the failure logic of the nodes within the graph, the algorithm determines whether the instantaneous failure of a node corresponds to a single component hard fault, the multi-node linkage offset corresponds to a multi-component coupled soft fault, or the slow drift of node parameters corresponds to a process parameter drift latent fault. Real-time detection values ​​of exhaust gas emissions are retrieved, the VOCs emission concentration offset is calculated, and a grading threshold interval matching algorithm is used to match the preset grading standard to complete the determination of the environmental impact level of the fault.

5. The fault diagnosis method for VOCs treatment equipment based on digital twins according to claim 1, characterized in that, In S4, the virtual-real two-way closed-loop correction simulation is divided into two fixed execution processes: forward backtracking simulation and reverse parameter correction. The forward backtracking simulation takes the fault characteristics as input and combines the graph path reverse traversal algorithm to traverse all related nodes in reverse along the directed graph of fault propagation to locate the original cause of the fault and sort out the fault propagation path step by step. The reverse parameter correction takes the real disturbance dynamic response data of the physical equipment as the benchmark and uses the gradient descent adaptive calibration algorithm to fine-tune the internal coupling coefficient of the twin in groups, and correct the multi-physics coupling parameters of the twin in batches to offset the inherent modeling error of the twin model.

6. The fault diagnosis method for VOCs treatment equipment based on digital twins according to claim 1, characterized in that, In S5, based on the two output dimensions of fault location results and fault environmental impact level, the hierarchical analysis decision algorithm is used to match and generate non-stop emergency control schemes and shutdown maintenance schemes respectively; a dedicated time series database is constructed, and disturbance dynamic response data, fault-related emission data, and model correction data are entered in a unified data format. Based on historical data, an incremental streaming learning algorithm is used to iteratively update the topology nodes of the fault propagation directed graph, and the internal coupling parameters of the twin simulation model are simultaneously optimized to achieve autonomous iteration of diagnostic capabilities.

7. A fault diagnosis system for VOCs treatment equipment based on digital twins, characterized in that, The system is used to perform the fault diagnosis method for VOCs treatment equipment based on digital twins as described in claim 1. The system includes: a multi-physics twin modeling synchronization module, a dual-mode parallel diagnosis module, a fault tracing and classification module, a virtual-real closed-loop simulation correction module, and a diagnostic data iteration module. The multi-physics twin modeling synchronization module is used to construct a full-domain multi-physics digital twin, establish a two-way data interaction link, and synchronously acquire real-time operating data of physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of the twin. The dual-mode parallel diagnostic module is used to run a passive residual monitoring and active micro-disturbance dual parallel diagnostic architecture, calculate the virtual and real dual-domain operating parameter residuals, separate non-fault interference residuals, obtain normal static residual data, send out controllable small-amplitude micro-disturbance signals and collect disturbance dynamic response data. The fault source tracing and classification module is used to integrate normal static residual data and disturbance dynamic response data, match the fault propagation directed graph, distinguish fault types, and determine the environmental impact level of the fault by associating VOCs emission concentration offset. The virtual-real closed-loop simulation correction module is used to perform virtual-real bidirectional closed-loop correction simulation, trace back the fault causes and propagation paths, and reverse correct the twin multi-physics coupling parameters. The diagnostic data iteration module is used to output fault diagnosis results and corresponding control and maintenance plans, complete the closed-loop storage of disturbance dynamic response data, fault-related emission data and model correction data, and update the fault map and twin simulation model.

8. The VOCs treatment equipment fault diagnosis system based on digital twin as described in claim 7, characterized in that, The system data flow is as follows: The multiphysics twin modeling synchronization module collects and synchronizes real-time operating data of the physical equipment, exhaust emission monitoring data, and fault-free standard simulation data of the twin, and transmits the synchronized full-domain data to the dual-mode parallel diagnostic module; After completing the calculation of virtual and real dual-domain operating parameter residuals, separation of non-fault disturbance residuals, output of normal static residual data, and acquisition of disturbance dynamic response data, the dual-mode parallel diagnostic module transmits the normal static residual data and disturbance dynamic response data to the fault source tracing and classification module; After completing the fault type differentiation and fault environmental impact level determination, the fault-related feature data is transmitted to the virtual and real closed-loop simulation correction module; After completing the fault path backtracking and twin parameter correction, the virtual and real closed-loop simulation correction module transmits the entire diagnostic process data to the diagnostic data iteration module; The diagnostic data iteration module completes the output of diagnostic results, data closed-loop storage, and update of fault maps and twin simulation models.

Images