Intelligent monitoring method and system for the operation status of ultrapure water treatment equipment
By injecting hydraulic step perturbation signals into an ultrapure water treatment unit, the viscoelastic response characteristics of the membrane module are collected and decoupled, solving the problem of not being able to distinguish between reversible fouling and irreversible degradation in existing technologies, and realizing precise operation and maintenance decisions and module protection.
Patent Information
- Application Number
- CN202611378021.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-07
- Publication Date
- 2026-11-17
AI Technical Summary
Existing status monitoring solutions for ultrapure water treatment equipment cannot effectively distinguish between reversible fouling and irreversible chemical degradation of membrane modules, leading to blind operation and maintenance decisions and potentially resulting in ineffective chemical cleaning or missing the optimal cleaning opportunity.
By injecting hydraulic step perturbation signals into the reverse osmosis high-pressure pump, acquiring multi-source transient sensing data, extracting the viscoelastic response feature matrix of the membrane module, decoupling and mapping the reversible fouling index and the irreversible aging index, and generating adaptive closed-loop control commands.
It enables precise identification of membrane module status and differentiated physical intervention, improving the accuracy of operation and maintenance decisions and avoiding unnecessary chemical cleaning and potential module damage.
Smart Images

Figure CN123102160A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of water treatment equipment operation and maintenance management technology, and more specifically, to an intelligent monitoring method and system for the operating status of an ultrapure water treatment device. Background Technology
[0002] Ultrapure water preparation systems are widely used in high-end industrial applications such as semiconductor manufacturing, optoelectronic chip cleaning, and precision electronic component production. Their core desalination units, reverse osmosis membrane modules, and continuous electrostatic desalination modules inevitably face performance degradation during long-term operation. Since the resistivity, particulate matter, and total organic carbon of the final ultrapure water have a direct and extremely sensitive impact on process yield, if abnormal degradation of the membrane module is not identified in time, it can lead to anything from the product water quality falling below the process tolerance threshold, resulting in batch product scrapping, to the entire production line being forced to shut down unplanned due to sudden breakthrough failure of the module, causing incalculable economic losses to high-value continuous production. Therefore, establishing an intelligent monitoring system capable of accurately sensing the deep operating status of the membrane module and adaptively generating maintenance decisions is of significant engineering practical importance for ensuring the long-term stable operation of ultrapure water preparation systems.
[0003] Current state monitoring schemes for ultrapure water treatment devices primarily rely on threshold judgment logic for macroscopic physical quantities under steady-state operating conditions. A typical approach involves continuously collecting parameters such as transmembrane pressure difference, permeate flow rate, and desalination rate during constant pressure or constant flow operation, comparing these absolute values with the initial operating baseline of the equipment, and triggering cleaning or alarms when the deviation exceeds a preset threshold. However, this technical approach has a fundamental flaw: it cannot distinguish whether the membrane module is currently experiencing reversible fouling that can be restored by chemical cleaning, or irreversible chemical degradation of the membrane substrate caused by residual free chlorine oxidation or water hammer impact, based solely on the decrease in flow rate and desalination rate. The underlying technical root of this predicament lies in the fact that existing monitoring methods acquire macroscopic readings under steady-state constant operating conditions. Under steady-state conditions, whether it is a viscoelastic gel layer formed by organic colloids or microbial adhesion, or a permanent expansion of rigid pore size caused by the breakage of polyamide crosslinks, both manifest as an increase in total thermodynamic resistance or an increase in solute permeability in the macroscopic mass transfer equation. The steady-state data itself lacks sufficient mathematical information to decouple and identify these two independent physical processes. Furthermore, existing technologies consistently lack the means to actively and dynamically detect the physical properties of the fouling layer. From a materials physics perspective, reversible fouling layers exhibit viscoelastic compressibility, while irreversible aging alters the intrinsic stiffness constant of the membrane substrate. The dynamic responses of these two to external transient stimuli are fundamentally different. However, because the system is always in a passive steady-state monitoring mode, this exploitable physical distinction has never been captured by existing solutions. These limitations ultimately lead to highly unpredictable operational decisions, either resulting in ineffective chemical cleaning of components that have already undergone irreversible degradation, thus accelerating secondary cleaning damage, or causing components in a reversibly fouled state to miss the optimal cleaning opportunity, evolving into stubborn fouling that is difficult to recover from.
[0004] Therefore, an optimized intelligent monitoring scheme for the operating status of ultrapure water treatment devices is desired. Summary of the Invention
[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and system for intelligent monitoring of the operating status of an ultrapure water treatment device.
[0006] According to one aspect of this application, a method for intelligent monitoring of the operating status of an ultrapure water treatment device is provided, comprising: S1, when the ultrapure water treatment device is in steady-state water production operation, a hydraulic step perturbation signal with a preset frequency amplitude is injected into the reverse osmosis high-pressure pump through the frequency conversion drive end, and multi-source transient sensor data is simultaneously collected at high frequency and clock-aligned and spliced to obtain transient hydraulic response time series data. S2, perform perturbation response dynamics feature extraction on transient hydraulic response time series data to obtain the viscoelastic response feature matrix of membrane module; S3, the membrane physical properties of the membrane module are decoupled and indexed by the membrane module viscoelastic response characteristic matrix to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. S4, a dual-track degradation state comprehensive evaluation of the component's reversible pollution index and irreversible aging index is conducted to obtain the dominant membrane state definition result. S5 optimizes the branching strategy based on the dominant membrane state definition results, the component reversible contamination index, and the component irreversible aging index to generate an adaptive closed-loop control instruction set, which is then sent to the programmable logic controller for execution.
[0007] According to another aspect of this application, an intelligent monitoring system for the operating status of an ultrapure water treatment device is provided, comprising: The hydraulic perturbation excitation injection and data acquisition module is used to inject a hydraulic step perturbation signal of preset frequency amplitude into the reverse osmosis high-pressure pump through the frequency conversion drive end when the ultrapure water treatment device is in steady-state water production operation. Simultaneously, it acquires multi-source transient sensor data at high frequency and performs clock alignment and splicing to obtain transient hydraulic response time series data. The perturbation response dynamics feature extraction module is used to extract perturbation response dynamics features from transient hydraulic response time-series data to obtain the viscoelastic response feature matrix of the membrane module. The membrane physical property decoupling and exponential mapping module is used to decouple and map the membrane physical properties of the membrane module viscoelastic response characteristic matrix to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. The dual-track degradation state comprehensive evaluation module is used to comprehensively evaluate the reversible pollution index and the irreversible aging index of the module to obtain the dominant membrane state definition result. The strategy optimization and control instruction generation module is used to perform branch strategy optimization on the dominant membrane state definition result, the component reversible pollution index and the component irreversible aging index to generate an adaptive closed-loop control instruction set, which is then sent to the programmable logic controller for execution.
[0008] Compared with existing technologies, this invention proposes an intelligent monitoring method for the operating status of an ultrapure water treatment device. It breaks through the traditional passive data reception mode under constant operating conditions, instead actively injecting a safe and controllable hydraulic step perturbation signal into the high-pressure variable frequency drive of the reverse osmosis system. This instantaneously excites the membrane module from steady-state equilibrium to a transient response state, while simultaneously acquiring dynamic evolution data of pressure, flow rate, and conductivity at high frequency. Because the reversible fouling layer has viscoelastic compressible properties while irreversible aging corresponds to the permanent expansion of rigid pores in the membrane substrate, their dynamic responses to transient pressure excitation differ fundamentally in terms of phase delay and mass transfer gradient. This solution utilizes this physical distinguishability, constructing a feature matrix through cross-correlation phase hysteresis analysis and transient response gradient extraction. This matrix is then mapped to the reversible fouling index and the irreversible aging index via a viscoelastic deformation model and a rigid expansion model, respectively, thereby achieving quantitative decoupling of the two degradation mechanisms in a purely software computational dimension. Based on this, the system evaluates the contribution ratio and determines the dominant cause of the dual-track index, and adaptively generates differentiated closed-loop control strategies accordingly, realizing a complete decision-making closed loop from state perception to physical intervention. Attached Figure Description
[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.
[0010] Figure 1 This is a flowchart of an intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application; Figure 3 This is a flowchart of step S3 of the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application; Figure 4 This is a flowchart of step S4 of the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application; Figure 5 This is a flowchart illustrating the operation status intelligent monitoring method for an ultrapure water treatment device according to an embodiment of this application. The flowchart compares the contribution ratio of pollution attenuation and the contribution ratio of aging attenuation with a preset absolute dominant boundary and completes enumeration classification inference to obtain the dominant membrane state definition result. Figure 6This is a block diagram of an intelligent monitoring system for the operating status of an ultrapure water treatment device according to an embodiment of this application. Detailed Implementation
[0011] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.
[0012] In the actual operation and maintenance of ultrapure water preparation systems, the root cause of membrane module performance degradation may be reversible surface fouling caused by the adhesion of colloids and microorganisms, or irreversible substrate degradation caused by residual chlorine oxidation or mechanical damage. Existing steady-state monitoring methods can only obtain the macroscopic appearance of overall resistance increase, and lack sufficient information to decouple and identify the two independent degradation mechanisms in a mathematical dimension, which leads to long-term blind operation and maintenance decisions. To address this issue, this solution involves actively injecting a hydraulic step perturbation signal of preset frequency amplitude into the reverse osmosis high-pressure pump via a frequency converter during the steady-state water production operation of the ultrapure water unit. This instantaneously excites the membrane module from a thermodynamic equilibrium state to a dynamic response state. Simultaneously, transient evolution data of feed water pressure, permeate flow rate, and permeate conductivity are captured at a millisecond sampling rate, and multi-channel clock alignment and splicing are completed. Subsequently, cross-correlation integration is performed on this transient time-series data to extract the phase hysteresis feature of the flow response relative to pressure excitation. The gradient feature of the ratio of conductivity change rate to pressure change rate is obtained through synchronous differential calculation. These two features are orthogonally fused into a viscoelastic response feature matrix for the membrane module. Furthermore, based on the viscoelasticity of the porous media... The elastic deformation model and the microporous rigid expansion model respectively decouple the physical properties of the feature matrix and independently quantify the reversible pollution index and irreversible aging index of the component. On this basis, the deviation of the dual-track index with its respective critical threshold is compared and the contribution ratio is normalized. The dominant membrane state is enumerated and classified by the decision tree model. Finally, based on the judgment result, the branch strategy is optimized by combining the two index values. The closed-loop control instruction set covering chemical cleaning triggering or blocking and frequency conversion PID compensation adjustment is adaptively generated and sent to the programmable logic controller for execution. Thus, the accurate identification of the deep degradation state of the membrane surface and the differentiated physical intervention are connected into a complete intelligent decision closed-loop link.
[0013] Figure 1 This is a flowchart of an intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application. Figure 1 and Figure 2As shown, the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application includes the following steps: S1, when the ultrapure water treatment device is in steady-state water production operation, a hydraulic step perturbation signal of preset frequency amplitude is injected into the reverse osmosis high-pressure pump through the frequency converter drive end, and multi-source transient sensor data is simultaneously acquired at high frequency and clock-aligned and spliced to obtain transient hydraulic response time-series data; S2, perturbation response dynamics feature extraction is performed on the transient hydraulic response time-series data to obtain the membrane module viscoelastic response feature matrix; S3, the membrane module viscoelastic response feature matrix is further processed... S4. Decoupling and index mapping of membrane physical properties to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate; S5. Performing a dual-track degradation state comprehensive evaluation on the module reversible fouling index and the module irreversible aging index to obtain the dominant membrane state definition result; S6. Optimizing the branching strategy on the dominant membrane state definition result, the module reversible fouling index, and the module irreversible aging index to generate an adaptive closed-loop control instruction set, which is then sent to the programmable logic controller for execution.
[0014] Specifically, in step S1, when the ultrapure water treatment device is operating in steady-state water production, a hydraulic step perturbation signal of preset frequency amplitude is injected into the reverse osmosis high-pressure pump through the frequency converter drive. Simultaneously, multi-source transient sensor data is acquired at high frequency and clock-aligned and stitched together to obtain transient hydraulic response time-series data. It should be noted that the multi-source transient sensor data includes the reverse osmosis feed water pressure signal, the product water flow rate signal, and the product water conductivity signal. It should be understood that when the ultrapure water treatment device performs conventional water production tasks in constant pressure or constant flow mode, whether the reverse osmosis membrane module surface is covered with a reversible fouling layer formed by organic colloids and microorganisms, or undergoes irreversible substrate degradation due to the breakage of polyamide crosslinks, the steady-state mass transfer equation will show a monotonically increasing total permeability resistance or a gradual decrease in solute rejection rate. Therefore, the pressure, flow rate, and conductivity data acquired under steady-state conditions do not mathematically possess sufficient freedom to decouple and identify these two independent degradation mechanisms. Therefore, in the technical solution of this application, when the ultrapure water treatment device is operating in steady-state water production, a hydraulic step perturbation signal of preset frequency amplitude is injected into the reverse osmosis high-pressure pump through the frequency converter drive end, and multi-source transient sensor data is simultaneously acquired at high frequency and clock-aligned and spliced to obtain transient hydraulic response time-series data. This instantaneously excites the membrane module from a thermodynamic equilibrium state to a dynamic response state, causing the reversible fouling layer to generate a nonlinear flow response with time delay characteristics under transient pressure excitation due to its viscoelastic compressibility properties. At the same time, the irreversible aging membrane surface generates a linear conductivity abrupt response without delay under the same excitation due to its rigid pore expansion properties. This creates unique kinetic fingerprint information for two types of degradation mechanisms in the time domain. In this way, without adding any additional physical sensor hardware, transient dynamic characteristic data far exceeding the dimension of steady-state monitoring information can be obtained only through the modification of software control logic, providing a sufficiently distinguishable original signal basis for the decoupling and identification of membrane physical properties in subsequent steps.
[0015] More specifically, in this embodiment, step S1 includes: after confirming that the ultrapure water treatment device is in steady-state water production operation, performing step signal modeling by superimposing a preset frequency offset on the high-pressure pump reference operating frequency data to obtain a step perturbation control signal; injecting the step perturbation control signal into the reverse osmosis high-pressure pump drive end through a frequency converter to excite transient hydraulic disturbance, and simultaneously opening an observation time window at the injection time to perform fixed-length sequence truncation of the reverse osmosis feed water pressure signal, product water flow signal, and product water conductivity signal to obtain a multi-source transient sensing dataset; performing equal-interval global clock resampling alignment and multi-channel column vector concatenation on each asynchronous sampling channel in the multi-source transient sensing dataset caused by hardware communication cycle differences to obtain transient hydraulic response time series data.
[0016] Specifically, in a specific example of this application, the implementation process of this step is as follows: First, the host computer software module of the monitoring system continuously reads the real-time values of the reverse osmosis feed water pressure signal, product water flow signal, and product water conductivity signal from the process image area of the programmable logic controller at a fixed scan cycle, and maintains a sliding time window of a preset length in the memory buffer. The window width of the sliding time window is set to 60 seconds, and the data sampling interval within the window is 100 milliseconds, thereby accumulating 600 continuous sampling points within the window at the end of each scan cycle. At the end of each scan cycle, the system calculates the statistical variance of the three signals within the sliding time window. Specifically, it calculates the variance of the 600 sampling points of the reverse osmosis feed water pressure signal relative to the window mean and compares the result with a preset pressure steady-state variance threshold. This pressure steady-state variance threshold is set based on the square of 0.5% of the rated operating pressure of the reverse osmosis system. In this example, under the condition that the operating pressure of the first-stage reverse osmosis is 1.5 MPa, this threshold is calibrated to 5.625 × 10⁻⁶. -5 The system calculates the window variance of the permeate flow rate signal and compares it with a steady-state variance threshold set at 1% of the rated permeate flow rate. It also calculates the window variance of the permeate conductivity signal and compares it with a steady-state variance threshold set at 2% of the steady-state permeate conductivity. The system determines that the ultrapure water treatment device is currently in a steady-state water production operation state only if the window variances of all three signals remain below their respective steady-state determination thresholds for three consecutive scan cycles, and writes the steady-state confirmation flag into the status register of the programmable logic controller. After the steady-state confirmation is completed, the system immediately reads the current output frequency value of the reverse osmosis high-pressure pump from the inverter's communication register via the Modbus TCP protocol as the reference operating frequency data of the high-pressure pump. In this example, the inverter is an ABB ACS510 series, and its output frequency data is stored in the holding register at address 40001. The data format is a 16-bit unsigned integer with a resolution of 0.01 Hz. The system reads the register value, converts it into the actual frequency value according to the scaling factor, and then stores it in the memory variable for later use.
[0017] Next, the system enters the step perturbation control signal generation stage. The system loads a preset frequency offset from the configuration parameter file. The principle for determining this preset frequency offset is that the transmembrane pressure change amplitude corresponding to the offset must meet two constraints: first, the pressure change amplitude is sufficient to excite a dynamic response on the membrane fouling layer that can be detected by the sensor resolution; second, the pressure change amplitude must not cause the conductivity of the permeate to momentarily exceed the upper limit of the allowable conductivity of the feed water of the continuous electro-desalination module. In the two-stage reverse osmosis ultrapure water system of this example, the rated operating frequency of the stage 1 reverse osmosis high-pressure pump is 45 Hz, corresponding to an operating pressure of 1.5 MPa. The linear mapping coefficient between frequency and pressure is calibrated on-site to be 0.033 MPa per Hz. The system sets the preset frequency offset to 0.5 Hz, corresponding to a transmembrane pressure perturbation amplitude of approximately 0.0165 MPa. This amplitude is approximately 1.1% of the rated operating pressure, and it has been verified that it will not cause the conductivity of the stage 2 reverse osmosis permeate to exceed the feed water limit of 5 micro-Siemens per centimeter. The system superimposes the high-pressure pump reference operating frequency data and the preset frequency offset according to the time-domain characteristics of the unit step function to generate a step perturbation control signal, the mathematical expression of which is:
[0018] in, The instantaneous frequency command value corresponding to the step perturbation control signal generated at time t, in Hertz; The reference operating frequency data of the high-pressure pump is read from the inverter communication register; in this example, its typical value is 45 Hz. The preset frequency offset is 0.5 Hz in this example; To set the trigger time A unit step function that jumps from 0 to 1 is used to precisely control the transient edge characteristics of sudden additions of perturbation signals. The time is determined by the system's internal real-time clock at the next whole second after steady-state confirmation is completed, to facilitate subsequent timestamp alignment. In this expression, This ensures that the perturbation signal is always superimposed based on the current actual operating conditions of the system rather than using a fixed absolute value, so that the perturbation amplitude can adaptively match the drift compensation of the high-pressure pump operating frequency after the change in raw water viscosity caused by the change in water temperature in different seasons. The step characteristic of the term ensures that the pressure excitation has a sufficiently steep rise edge, making its rise time much shorter than the viscoelastic relaxation time constant of the fouling layer, thereby exciting a complete creep response process on the fouling layer for subsequent feature extraction steps to capture. After completing signal modeling, the system encodes the step perturbation control signal according to the data format of the inverter communication protocol. Specifically, the target frequency value is multiplied by 100, rounded, and written into the inverter's frequency setpoint register address 40002. After receiving the new frequency setpoint value, the inverter performs a frequency jump within its internal acceleration / deceleration time setting range. In this example, the acceleration / deceleration time is pre-configured to 0.1 seconds to ensure that the steepness of the frequency jump edge meets the ideal characteristics of the step signal.
[0019] Then, within the same program scan cycle of writing the step perturbation control signal into the inverter's frequency setpoint register, the system synchronously opens the observation time window through the high-speed counter interrupt mechanism of the programmable logic controller. The delay between the start time of this observation time window and the time of writing into the frequency setpoint register is controlled to be within one PLC scan cycle. In this example, the typical scan cycle of the Siemens S7-1200 series PLC is 10 milliseconds, which is negligible compared to the hydraulic time constant of the membrane module. The duration of the observation time window is determined based on the hydraulic response characteristics of the membrane module, specifically set to 3 to 5 times the hydraulic time constant of the membrane module. This hydraulic time constant is determined by the membrane housing volume, the number of membrane elements, and the influent flow rate. In this example, the 8-inch 6-core membrane housing system was calibrated to approximately 1.2 seconds through on-site step response testing. Therefore, the duration of the observation time window is set to 5 seconds. During the observation window, the system synchronously acquires three sensor signals at a sampling rate of 1000 Hz. Specifically, the reverse osmosis feed water pressure signal is generated by a pressure transmitter installed on the first-stage reverse osmosis feed water header, which outputs a 4 to 20 mA current signal. This signal is then converted from analog to digital by the PLC's high-speed analog input module at a 1-millisecond sampling interval and written to the ring buffer. The product water flow signal is generated by an electromagnetic flowmeter installed on the first-stage reverse osmosis product water header, which outputs a 4 to 20 mA current signal. This signal is acquired by another channel of the same high-speed analog input module at the same sampling interval. The product water conductivity signal is generated by a conductivity meter installed on the first-stage reverse osmosis product water pipeline, which outputs the measured value via an RS485 digital bus at its highest refresh rate. In this example, the GF Signet 3-9900 series conductivity meter can be selected, with a maximum data refresh rate of 20 times per second. Within a 5-second observation window, 5000 sampling points are extracted from the pressure channel and the flow channel, and 100 sampling points are extracted from the conductivity channel. The above three fixed-length sequences, together with their respective local timestamps, are encapsulated into a multi-source transient sensing dataset and stored in the memory data structure of the host computer.
[0020] Furthermore, because the three signals in the multi-source transient sensing data arrive at the data acquisition end via different physical transmission media and communication protocols, although the pressure and flow channels are acquired through the same analog input module, there is an inherent sequential delay on the order of microseconds between the channels within the module. The conductivity channel, transmitted via the RS485 digital bus, has a communication cycle delay on the order of tens of milliseconds. The original timestamps of the three signals do not have a strict synchronous alignment relationship. Directly analyzing the data from each channel in parallel according to their index will introduce false time offset features. The system performs equidistant global clock resampling and alignment processing to address the above asynchronous sampling problem. Specifically, the system first establishes a global synchronous time axis in memory with a fixed step size of 1 millisecond. The starting time of this time axis is aligned to the trigger time of the observation time window. The termination time is Adding 5 seconds, the axis contains a total of 5001 equally spaced time nodes. For the pressure and flow channels with a sampling rate of 1000 Hz, the system performs nearest neighbor matching between the local timestamps of their original sampling points and the global time axis nodes. Sub-millisecond offsets caused by internal channel scanning delays are corrected using a linear interpolation algorithm; that is, for any target time node on the global time axis... Locate the corresponding channel in the original sequence that satisfies For adjacent frames of a given relationship, the virtual sampled value of that node is calculated using a linear weighted average based on the time distance. For the conductivity channel with a sampling rate of only 20 Hz, since its original sampling interval of 50 milliseconds is much larger than the 1 millisecond step of the global time axis, the system uses a cubic spline interpolation algorithm to upsample and reconstruct its 100 original sampling points, expanding them to 5001 interpolated data points that correspond one-to-one with the global time axis. Cubic spline interpolation ensures data continuity while avoiding the sawtooth-like pseudo-fluctuations that linear interpolation might introduce under sparse sampling conditions. After clock resampling and alignment of all three signals, the system uses the aligned pressure sequence as the first column vector, the flow rate sequence as the second column vector, and the conductivity sequence as the third column vector, concatenating them along the column dimension according to the same timestamp index to construct a structured matrix with 5001 rows and 3 columns. This matrix represents the transient hydraulic response time series data. Each row of the transient hydraulic response time-series data corresponds to one time node on the global time axis. The three columns respectively carry the transient evolution trajectories of the inlet pressure, product water flow rate, and product water conductivity of the membrane module under perturbation excitation. The strict alignment of the three signals in the time dimension ensures that the phase relationship and rate of change ratio extracted by subsequent cross-correlation and differentiation operations have physical authenticity and are not pseudo-correlation introduced by communication delay. After the perturbation flaw detection cycle is completed, the system restores the original high-pressure pump reference operating frequency data to the frequency setpoint register through the frequency converter communication interface, so that the reverse osmosis system returns to the steady-state water production operation state. The entire perturbation flaw detection process lasts for about 5 to 6 seconds and has no detectable impact on the continuous water production capacity and terminal water quality stability of the ultrapure water system. The system repeats the above process according to the preset flaw detection execution cycle, which is set to be executed once every 4 hours in this example. Each execution outputs a new set of transient hydraulic response time-series data for subsequent perturbation response kinetic feature extraction steps.
[0021] Specifically, in step S2, perturbation response dynamics features are extracted from the transient hydraulic response time-series data to obtain the viscoelastic response feature matrix of the membrane module. It should be understood that although the transient hydraulic response time-series data output in the previous step has fully captured the dynamic evolution trajectory of the membrane module after perturbation excitation in the time dimension, this data is still the original three-channel time-domain waveform signal. The transient changes in pressure, flow rate, and conductivity are superimposed in the same time-series matrix in a hybrid coupling form, and have not yet been transformed into a structured feature representation that can directly characterize the differences in the physical state of the membrane surface. If the original time-domain data is directly input into the subsequent physical model mapping step, the model will be unable to perform effective decoupling operations due to the lack of a clear feature input dimension. Therefore, in the technical solution of this application, perturbation response dynamic features are further extracted from the transient hydraulic response time-series data to obtain the viscoelastic response feature matrix of the membrane module. This extracts two types of dynamic fingerprints with clear physical correspondence from the original time-domain waveform. The phase hysteresis feature is used to characterize the viscoelastic compressible dynamic properties of the membrane surface adhesion layer, and the transient response gradient feature is used to characterize the rigid mass transfer properties of the micropores in the membrane substrate. The two types of features are then structurally fused in an orthogonal space. In this way, the high-dimensional original transient time-domain signal can be compressed into a low-dimensional but highly discriminative physical feature matrix, enabling the subsequent membrane physical property decoupling step to perform targeted model mapping operations for the two independent degradation mechanisms of reversible fouling and irreversible aging in a defined feature space.
[0022] More specifically, in this embodiment, step S2 includes: performing cross-correlation phase hysteresis feature calculation and data demultiplexing on the transient hydraulic response time series data to obtain a phase hysteresis feature sequence and residual pressure conductivity time series data; performing synchronous instantaneous rate of change differentiation on the high-frequency pressure sequence and high-frequency conductivity sequence in the residual pressure conductivity time series data and calculating the ratio sequence of conductivity rate of change to pressure rate of change to obtain a transient response gradient feature sequence; and orthogonally concatenating the phase hysteresis feature sequence and the transient response gradient feature sequence in the column dimension to obtain the viscoelastic response feature matrix of the membrane module.
[0023] Specifically, in a concrete example of this application, firstly, cross-correlation phase hysteresis characteristic calculation and data demultiplexing are performed on the transient hydraulic response time series data. The system extracts the pressure sequence from the first column and the flow sequence from the second column of the three-column structure of the transient hydraulic response time series data. Cross-correlation integration is then performed on these two signals within a sliding time window framework to extract the phase delay of the flow response relative to the pressure excitation. Specifically, the system segments the total sequence of 5001 time nodes according to a preset sliding window width. This sliding window width is set to 500 sampling points, corresponding to a time span of 500 milliseconds, and the sliding step size is set to 100 sampling points, or 100 milliseconds. This generates a total of 46 cross-correlation calculation windows within the entire 5-second observation time window. Within each sliding window, the system performs a normalized cross-correlation function calculation on the pressure and flow sequences. It searches for the offset corresponding to the maximum value of the cross-correlation function within a preset search range, defined as 0 to 200 milliseconds, to match the typical timescale of the viscoelastic relaxation process of the membrane fouling layer. The offset corresponding to the maximum value is the phase delay time of the flow rate relative to the pressure within that window. The mathematical expression for this cross-correlation extremum optimization is:
[0024] in, The maximum phase delay time of the flow response relative to the pressure excitation, calculated within the k-th sliding window, is expressed in milliseconds and constitutes the k-th element in the output phase hysteresis feature sequence; k is the index of the sliding window, ranging from 1 to 46. The time offset variable is used to iterate through the cross-correlation function for searching. The upper bound for the preset offset search is set to 200 milliseconds in this example; The first column of the transient hydraulic response time series is the pressure sequence at the start time of the k-th window. Pressure sample value after offset by n sampling steps; Add offsets to the second column of the transient hydraulic response time series at the corresponding time points. The subsequent flow sampling value; This is the number of sampling points within a single sliding window, which is 500 in this example; The sampling step size for the global synchronization time axis is 1 millisecond in this example. In this expression, the point-by-point multiplication accumulation in the numerator characterizes the waveform similarity between the pressure excitation signal and the flow response signal at a specific time offset, while the dual-path root-mean-square normalization in the denominator eliminates the interference of signal amplitude differences on the accuracy of phase estimation. The operator then locates the optimal delay time point from all candidate offsets that causes the normalized cross-correlation coefficient to peak. From the perspective of membrane physics, when a viscoelastic compressible organic gel layer or microbial membrane is attached to the surface of the membrane module, a step pressure excitation requires first compressing this attached layer to shrink its volume before the pressure can be transmitted to the permeate side of the membrane to induce a flow change. This creep compression process manifests in the time domain as a detectable millisecond-level delay in the flow response relative to the pressure excitation. However, when the membrane is healthy or undergoes only irreversible aging with rigid pore size expansion, the pressure excitation passes directly through the membrane surface with almost no delay, inducing an instantaneous flow response. Therefore, The magnitude of this value directly reflects the thickness and viscoelastic stiffness characteristics of the reversible fouling layer on the membrane surface. After performing the above calculations for each of the 46 windows, the system then... A 46-column phase hysteresis feature sequence is generated by arranging the data in chronological order. Simultaneously, during the cross-correlation calculation, the system performs data demultiplexing, extracting the first column (pressure sequence) and the third column (conductivity sequence) from the transient hydraulic response time-series data and repackaging them into pressure-conductivity residual time-series data. This data structure retains the original 5001-row time node length and 1-millisecond sampling interval, with the first column being the pressure sequence and the second column being the conductivity sequence, and is directly passed through to the next calculation stage.
[0025] Next, the system performs synchronous instantaneous rate-of-change derivative and ratio sequence calculation on the residual time-series data of pressure conductivity to extract transient response gradient features. The system reads the pressure sequence and conductivity sequence from the residual time-series data of pressure conductivity, and performs first-order forward differential operations on the two sequences at the same time nodes based on the finite difference principle, calculating the instantaneous rate of change at each time node with a sampling interval of 1 millisecond as the differential step size. Subsequently, the system calculates the ratio of the instantaneous rate of change of conductivity to the instantaneous rate of change of pressure at each time node, generating an original gradient ratio sequence of length 5000. To maintain consistency with the phase hysteresis feature sequence on the time scale, the system performs piecewise root mean square statistics on the original gradient ratio sequence using the same sliding window configuration as the cross-correlation calculation, that is, calculating the root mean square energy value of the gradient ratio within each 500-millisecond wide window, thereby compressing the 5000-point original sequence into 46 window-level statistics, forming a transient response gradient feature sequence of length 46. The above gradient ratio calculation is expressed as:
[0026] in, For discrete time nodes The transient response gradient value of conductivity to pressure calculated at the point; and Do not refer to the second column of the residual time series data of pressure conductivity at the current time node. Compared with the previous time point The sampled values are in microsiemens per centimeter. and These are the sampled values of the pressure sequence in the first column of the residual time series data of pressure conductivity at the corresponding time nodes, with the unit being megapascals; This is a dimensionless engineering scaling factor, set to 1000 in this example. It scales the gradient value from the micro-Siemens per centimeter per megapascal range to a suitable range for subsequent matrix operations, while preventing numerical overflow that might occur when the pressure difference fraction in the denominator is close to zero due to small perturbation amplitudes. In this expression, the numerator part... The captured value is the instantaneous jump in product water conductivity caused by perturbation pressure changes. This jump directly reflects the instantaneous response sensitivity of transmembrane solute transport flux under pressure excitation; the denominator part... The pressure driving force increment that causes the aforementioned change in conductivity; the ratio of the two. In a physical sense, this is equivalent to the differential solute permeation coefficient of the membrane module at that moment. When irreversible aging of the membrane substrate leads to a rigid expansion of the micropores in the polyamide desalination layer, the same magnitude of pressure perturbation will cause a significant increase in solute permeation, manifested as... The numerical value increases systematically; however, when only reversible surface fouling exists on the film surface and the substrate structure is intact, the additional resistance of the fouling layer to solute transport actually makes the film surface more susceptible to fouling. The system maintains a low level or exhibits nonlinear fluctuations due to viscoelastic compression. After completing root mean square statistics for 46 windows, the system outputs a transient response gradient feature sequence.
[0027] Then, Gaussian smoothing and noise reduction processing is performed on the phase hysteresis feature sequence and the transient response gradient feature sequence. Since the cross-correlation extremum optimization operation and the finite difference derivative operation are both sensitive to high-frequency random noise in the original signal, the eigenvalues of individual windows may exhibit occasional spikes due to noise interference. The system applies 1D Gaussian smoothing filtering to the two feature sequences of length 46 in the time dimension. The radius of the single-sided window of the Gaussian kernel is set to 3, that is, the total width of the convolution kernel is 7 window nodes, and the standard deviation parameter is set to 1.5. After Gaussian convolution, the occasional noise spikes in the two sequences are effectively suppressed while the overall trend characteristics of the signal are preserved.
[0028] Finally, the Gaussian-smoothed phase hysteresis feature sequence and the transient response gradient feature sequence are orthogonally concatenated along the column dimension to construct the viscoelastic response feature matrix of the membrane module. Specifically, the system uses the smoothed phase hysteresis feature sequence as the first column vector and the smoothed transient response gradient feature sequence as the second column vector, and concatenates them side-by-side along the column dimension according to the same window index to construct a structured matrix with dimensions of 46 rows by 2 columns. This matrix is the viscoelastic response feature matrix of the membrane module. The first column of this viscoelastic response feature matrix carries the evolution information of the viscoelastic compressible dynamic properties of the membrane surface adhesion layer in the time dimension, and its numerical amplitude is positively correlated with the thickness and stiffness of the reversible fouling layer; the second column carries the evolution information of the rigid mass transfer properties of the micropores of the membrane substrate in the time dimension, and its numerical amplitude is positively correlated with the degree of pore size expansion caused by irreversible aging. The orthogonal arrangement of the two columns of features in the column dimension ensures that the subsequent decoupling steps can complete the independent extraction of the two types of physical properties through simple column index slicing operations, without the need to perform additional dimension decomposition algorithms. The viscoelastic response characteristic matrix of the membrane module is the sole output of this step and is directly supplied to the subsequent steps of membrane physical property decoupling and exponential mapping for processing.
[0029] Specifically, in step S3, the membrane module viscoelastic response feature matrix is decoupled from its membrane physical properties and subjected to exponential mapping to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. It should be understood that although the membrane module viscoelastic response feature matrix output in the previous step has organized the dynamic fingerprints of the two types of physical properties into a unified matrix structure in the form of orthogonal column vectors, the values in this matrix are still raw engineering feature quantities in units of millisecond-level delay time and dimensionless gradient ratio. They have not yet been transformed into a normalized health state scale with clear physical degradation semantics. Therefore, the operation and maintenance decision-making system cannot directly determine the current degree of reversible fouling accumulation or the stage of irreversible aging of the membrane module based on the absolute magnitude of the raw feature values. Therefore, in the technical solution of this application, the membrane module viscoelastic response feature matrix is further decoupled from its membrane physical properties and subjected to exponential mapping to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. This allows the original kinetic features in the orthogonal feature space to be mapped onto a continuous health evaluation scale from 0 to 100 via their respective corresponding physical degradation mechanism models, enabling a quantitative characterization of the severity of the two independent degradation channels with unified dimensions and comparability. In this way, abstract signal-level feature values can be transformed into scalar indices with intuitive physical degradation semantics, providing a foundation of input data that can be directly compared with critical thresholds for subsequent comprehensive assessment of dual-track attenuation states.
[0030] Figure 3This is a flowchart of step S3 of the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application. Figure 3 As shown, step S3 includes: S31, based on the feature channel indexing mechanism, orthogonally slicing and decoupling the viscoelastic response feature matrix of the membrane module in the column dimension to obtain the first channel data characterizing the dynamic delay characteristics and the second channel data characterizing the mass transfer agility characteristics, thereby obtaining the smooth phase hysteresis feature set and the smooth response gradient feature set; S32, performing porous medium viscoelastic deformation mapping and reversible contamination quantization on the smooth phase hysteresis feature set to obtain the module reversible contamination index; S33, performing micropore rigid expansion model mapping and irreversible aging quantization on the smooth response gradient feature set to obtain the module irreversible aging index.
[0031] More specifically, in a concrete example of this application, firstly, an orthogonal slicing decoupling operation is performed on the viscoelastic response feature matrix of the membrane module in the column dimension based on the feature channel indexing mechanism. By receiving the 46-row, 2-column viscoelastic response feature matrix of the membrane module from the previous step, it is sliced according to the column index number. The first column vector is extracted as the first channel data characterizing the dynamic delay characteristics, and the second column vector is extracted as the second channel data characterizing the mass transfer agility characteristics. Since the matrix has already undergone Gaussian smoothing and noise reduction processing in the previous step, the first channel data is the smoothed phase hysteresis feature set, and the second channel data is the smoothed response gradient feature set. Each of the two feature sets contains 46 time-series window-level statistics, respectively carrying the delay response intensity information of the viscoelastic compressible layer of the membrane surface under perturbation excitation and the solute mass transfer sensitivity information of the micropores of the membrane substrate under perturbation excitation. It is worth mentioning that these two columns of features have been deterministically orthogonally arranged in the column dimension according to different physical dimensions during the previous step's construction process; therefore, this orthogonal slicing operation can be completed directly through column index addressing.
[0032] Next, porous media viscoelastic deformation mapping and reversible fouling quantification operations are performed on the smooth phase hysteresis feature set to obtain the module reversible fouling index. From the perspective of materials physics in membrane separation engineering, the organic colloidal layer, microbial membrane, or inorganic gel layer attached to the reverse osmosis membrane surface is structurally equivalent to a porous viscoelastic solid. When a step pressure is applied to this attached layer, the internal pore liquid is forced out, triggering skeletal creep and contraction. This process manifests as a delay in flow response in the time domain, and the magnitude of the delay time is directly positively correlated with the total thickness and bulk modulus of the attached layer. The system performs window mean statistical calculation on 46 window-level delay time values in the smooth phase hysteresis feature set to calculate the overall energy level of phase delay within the flaw detection cycle. Subsequently, this mean statistic is substituted into a nonlinear saturated mapping model based on the stretching exponential function for normalization and quantification to generate a module reversible fouling index with a value between 0 and 100. The mathematical expression of this mapping model is:
[0033] in, The calculated component reversible fouling index is a continuous scale from 0 to 100, where 0 indicates that there is no detectable fouling on the membrane surface and 100 indicates that the fouling layer thickness has reached the limit state of completely blocking the transmembrane flow channel; N is the total number of effective windows in the smooth phase hysteresis feature set, which is 46 in this example. This represents the smoothed phase delay time value of the k-th window in the smoothed phase hysteresis feature set, in milliseconds. For viscoelastic damping empirical conversion constant, its physical meaning is the scaling factor that maps the delay time in milliseconds to the dimensionless fouling accumulation intensity. This constant is determined by the comparative calibration test of the membrane module at the time of manufacture under the conditions of a clean membrane surface and a standard fouling layer of known thickness. In the polyamide reverse osmosis membrane system of this example, the value is 0.08 per millisecond. The stretching exponential shape factor controls the linear response sensitivity of the mapping curve in the low-pollution region and the saturation compression degree in the high-pollution region; in this example, it is set to 1.5. In this expression, The arithmetic mean of the delay times of the 46 windows is calculated, compressing the window-level temporal fluctuations into a single statistic that characterizes the overall contamination level of this flaw detection cycle. The product with this mean transforms the millisecond-level physical delay time into a dimensionless intermediate variable characterizing the intensity of fouling accumulation; the outer layer stretching exponential function It provides a monotonically increasing saturated mapping characteristic from 0 to 1, ensuring that the index value increases approximately linearly with delay time in the low-pollution range to provide sufficient monitoring sensitivity, while the index growth rate naturally decays in the high-pollution range to avoid spurious jumps in the index value caused by amplified delay time measurement noise. The outermost layer is multiplied by 100 to scale the normalized ratio value to a percentage scale for easy understanding by engineers. From the perspective of operation and maintenance practice of ultrapure water preparation systems, when the reversible fouling index of this component is in the range of 0 to 30, it indicates that the membrane surface is in a normal and clean operating state; the range of 30 to 60 indicates that reversible fouling has begun to accumulate but has not yet affected the quality of the produced water; and the range of 60 to 100 indicates that the fouling layer has reached the level requiring in-situ chemical cleaning intervention.
[0034] Then, the micropore rigid expansion model mapping and irreversible aging quantification operation are performed on the smooth response gradient feature set to obtain the irreversible aging index of the module. From the perspective of membrane substrate degradation mechanism, when the desalination layer of polyamide composite membrane undergoes polymer crosslinking bond breakage due to free chlorine oxidation attack or water hammer mechanical impact, the micropores on the membrane surface undergo irreversible rigid expansion. Under the same pressure perturbation, the solute permeation is systematically increased compared to the healthy membrane surface, manifested as a continuous increase in the transient response gradient feature value. The system performs root mean square energy calculation on 46 window-level gradient statistics in the smooth response gradient feature set to extract the overall offset energy level of the gradient signal within the flaw detection cycle. Subsequently, this root mean square value is substituted into the micropore rigid expansion classification model based on the Sigmoid activation function for normalization and quantization. This model uses the healthy baseline permeation gradient threshold calibrated at the time of membrane module delivery as the reference zero point, and performs nonlinear amplification mapping on the part exceeding the baseline threshold to generate the module irreversible aging index with a value between 0 and 100. The mathematical expression of this mapping model is:
[0035] in, The irreversible aging index of the component is calculated, and its value ranges from 0 to 100 on a continuous scale. 0 indicates that the microstructure of the membrane substrate is completely healthy, and 100 indicates that the desalination layer has undergone severe chemical degradation or physical tearing, resulting in the loss of solute retention capacity. N is the total number of effective windows in the smooth response gradient feature set, which is 46 in this example. For the smooth transient response gradient statistics of the k-th window in the smooth response gradient feature set; The rated permeation gradient rigid threshold constant is determined by calibration tests under factory health baseline conditions for the membrane module. This threshold represents the normal solute permeation gradient level that a healthy membrane surface should exhibit under the same perturbation excitation conditions. In this example, it is calibrated to 2.5 by the standardized step test when the membrane element leaves the factory. This is the damage activation sensitivity gain factor, which controls the steepness of the rate at which the exponential value climbs towards a severely aged state after the actual root mean square gradient value exceeds a baseline threshold. In this example, it is set to 3.0. In this expression, The root mean square (RMS) operation is performed on 46 window-level gradient values, compressing temporal fluctuations into a single energy scalar characterizing the overall shift in solute mass transfer sensitivity of the membrane surface within the flaw detection cycle. Compared to the arithmetic mean, the RMS operator is more effective at capturing anomalous high gradient mutations occurring in individual windows; subtracting... The operation sets the healthy baseline state to the center zero of the Sigmoid function, so that when the gradient energy level is equal to the factory calibration value, the exponent is exactly at the neutral position of 50. The linear scaling factor, which is the independent variable of the Sigmoid function, determines the width of the transition range from the low aging state to the high aging state. The larger the value, the steeper the transition and the more decisive the judgment. From the perspective of the whole life cycle management of reverse osmosis membrane modules, the upper and lower saturation characteristics of the Sigmoid function ensure that the exponent value stably approaches 0 under healthy membrane conditions without producing false positive values due to measurement noise, and stably approaches 100 under severe degradation conditions without diverging without bounds due to the continuous increase of the gradient value.
[0036] Furthermore, after completing the two independent model mapping calculations mentioned above, the component reversible contamination index and the component irreversible aging index are packaged as the final outputs of this step. The component reversible contamination index is a scalar value ranging from 0 to 100, quantitatively characterizing the degree of reversible water permeability resistance accumulation on the surface of the membrane module due to the adhesion of organic colloids, microbial films, or inorganic scaling during the current flaw detection cycle. The component irreversible aging index is a scalar value ranging from 0 to 100, quantitatively characterizing the degree of irreversible desalination capacity loss of the membrane substrate due to chemical oxidation degradation or mechanical damage during the current flaw detection cycle. The uniformity of the numerical dimensions of the two indices ensures that the subsequent dual-track degradation state comprehensive assessment step can perform a fair horizontal comparison and contribution ratio quantification calculation of the severity of the two independent degradation channels in the same coordinate system, without performing additional dimension conversion or normalization preprocessing. The system writes the two scalar indices, along with the timestamp of the current flaw detection cycle, into the operating status database for persistent storage to support the tracking of degradation trends throughout the entire life cycle. At the same time, the latest component reversible contamination index and component irreversible aging index are directly transmitted to the subsequent dual-track degradation status comprehensive assessment step for consumption.
[0037] Specifically, in step S4, a dual-track degradation state comprehensive assessment is performed on the component's reversible contamination index and irreversible aging index to obtain the dominant membrane state definition result. It should be understood that although the component's reversible contamination index and irreversible aging index output in the previous step have quantified the severity of the two independent degradation channels onto a unified scale of 0 to 100, the absolute values of the two indices, existing independently, have not yet been transformed into classification judgment conclusions that can directly drive the selection of operation and maintenance decision branches. The operation and maintenance system cannot directly determine which degradation mechanism is dominant in the current membrane module performance degradation based solely on two independent scalar values, nor can it directly choose whether to perform chemical cleaning or initiate the module replacement process. Therefore, in the technical solution of this application, a dual-track degradation state comprehensive assessment is further performed on the component's reversible contamination index and irreversible aging index to obtain the dominant membrane state definition result. This allows for the normalization and quantification of the deviation degree of each of the two degradation channels from the safe operating boundary within a unified assessment framework, and the deterministic classification inference of dominant attribution is completed based on the logical comparison of the contribution ratio and the preset judgment boundary. In this way, the health index of two continuous numerical domains can be transformed into discrete enumeration classification labels with clear operation and maintenance strategy orientations. This allows the subsequent closed-loop control instruction generation steps to directly select the corresponding strategy branch based on the classification label without having to perform complex multi-condition combination judgments in the continuous numerical space.
[0038] Figure 4 This is a flowchart of step S4 of the intelligent monitoring method for the operating status of an ultrapure water treatment device according to an embodiment of this application. Figure 4 As shown, step S4 includes: S41, comparing the component's reversible contamination index and irreversible aging index with the preset reversible contamination cleaning threshold and irreversible chemical failure threshold in the system knowledge base to obtain the reversible physical resistance deviation and the irreversible chemical damage deviation; S42, quantifying the two-dimensional attenuation contribution ratio of the reversible physical resistance deviation and the irreversible chemical damage deviation to obtain the contamination attenuation contribution ratio and the aging attenuation contribution ratio; S43, based on the operation and maintenance strategy classification decision tree model, comparing the contamination attenuation contribution ratio and the aging attenuation contribution ratio with the preset absolute dominant boundary and completing the enumeration classification inference to obtain the dominant membrane state definition result.
[0039] Accordingly, in step S41, the reversible fouling index and the irreversible aging index of the component are compared with the preset reversible fouling cleaning threshold and irreversible chemical failure threshold in the system knowledge base to obtain the deviation of reversible physical resistance and the deviation of irreversible chemical damage. It should be understood that since the reversible fouling index and the irreversible aging index of the component output in the previous step vary independently on a continuous scale from 0 to 100, not every non-zero value of the index indicates that the membrane module is in an abnormal state requiring intervention. In the actual operation of the ultrapure water preparation system, there is a trace amount of initial adhesion on the membrane surface at the beginning of operation, and the membrane substrate itself has an inherent microporous distribution. These factors will cause the two indices to show non-zero background base values when the membrane module is in the normal healthy operating range. If any non-zero value of the index is regarded as a degradation contribution, it will cause an overreaction to normal fluctuations within the healthy range. Therefore, in the technical solution of this application, the reversible contamination index and the irreversible aging index of the component are compared with the preset reversible contamination cleaning critical threshold and irreversible chemical failure critical threshold in the system knowledge base to obtain the reversible physical resistance deviation and the irreversible chemical damage deviation. This establishes a deviation coordinate system with their respective critical thresholds as zero reference points. Only the positive over-limit increment exceeding the safe operating boundary is extracted as the effective attenuation contribution signal, while normal background fluctuations within the safe range are automatically truncated to zero. This effectively eliminates the dilution interference of the index base value within the healthy operating range on the subsequent contribution ratio quantification calculation, ensuring that the attenuation attribution determination in subsequent steps is based solely on the effective degradation increment that truly exceeds the safe boundary, avoiding unnecessary maintenance interventions triggered by background noise when the membrane module is in normal service.
[0040] More specifically, in a concrete example of this application, firstly, two critical threshold parameters matching the currently online membrane module model are loaded from the configuration parameter table of the local knowledge base. The reversible fouling cleaning critical threshold is calibrated based on the membrane element manufacturer's guidelines in the factory technical manual that chemical cleaning should be performed when the permeate flux declines to 85% of its initial value. This is combined with the statistical average of the reversible fouling index of the module before each chemical cleaning, recorded in historical on-site operation and maintenance data. The smaller of these two values is taken as the final threshold to ensure the conservatism of the intervention timing. In this example, the threshold for the Hydranautics PROC10 polyamide composite reverse osmosis membrane system is calibrated to 55. The irreversible chemical failure critical threshold is calibrated based on the irreversible aging index value of the module when the desalination rate drops below 96% during accelerated life testing. This 96% desalination rate lower limit is determined by reverse calculation based on the specification requirement of the ultrapure water system for a stage 2 reverse osmosis permeate conductivity not exceeding 5 microSiemens per centimeter. In this example, this threshold is calibrated to 40 based on accelerated test data. The system loads the two thresholds into memory variables for later use. During the loading process, threshold rationality checks are performed to ensure that the critical threshold for reversible contamination cleaning is within the legal range of 30 to 80 and the critical threshold for irreversible chemical failure is within the legal range of 20 to 70.
[0041] Next, a positive over-limit comparison calculation is performed on the component's reversible fouling index. The system reads the current value of the component's reversible fouling index from the previous step, performs an arithmetic subtraction operation with it to obtain the difference, and then applies a linear rectification operation to this difference. When the difference is positive, it is directly retained as the deviation output; when the difference is negative or zero, the deviation is set to zero. The physical meaning of the linear rectification operation is: when the component's reversible fouling index has not yet reached the cleaning critical threshold, although there may be a certain degree of deposit accumulation on the membrane surface, this accumulation level is still within the safe operating range that does not affect the system's product water quality and flux specifications. At this time, the reversible fouling channel should not contribute to the attribution determination of system performance degradation. Similarly, the system performs the same positive over-limit comparison calculation on the component's irreversible aging index, calculates the difference with the irreversible chemical failure critical threshold, and applies linear rectification to obtain the irreversible chemical damage deviation. The mathematical expression of the above positive over-limit comparison is:
[0042] in, The calculated reversible physical resistance deviation is characterized by the positive over-limit increment of the component's reversible contamination index after it exceeds the critical safety boundary for cleaning. The value ranges from 0 to 45, which is the maximum possible deviation of the component's reversible contamination index from the threshold of 55 to the full scale of 100. The calculated irreversible chemical damage deviation is characterized by the positive over-limit increment of the irreversible aging index of the component after it exceeds the critical safety boundary of failure. The value range is 0 to 60, which is the maximum possible deviation of the irreversible aging index of the component from the threshold of 40 to the full scale of 100. This is the current flaw detection cycle value of the component reversible contamination index passed in the previous step; This is the current flaw detection cycle value of the irreversible aging index of the component passed in the previous step; The critical threshold for reversible contamination cleaning, loaded from the system knowledge base, is set to 55 in this example. The critical threshold for irreversible chemical failure, loaded from the system knowledge base, is set to 40 in this example. This is a linear rectification operator that truncates the negative range to zero while maintaining the identity mapping for the positive range. In this expression, and The subtraction operation establishes the respective critical thresholds as the zero reference origin deviating from the coordinate system, so that the subsequent contribution ratio calculation is performed in the relative offset space based on the critical threshold rather than in the absolute exponential space from 0 to 100. This coordinate transformation ensures that only the effective degradation increment that truly exceeds the safety boundary is included in the denominator basis of the decay attribution. The linear rectification characteristic provides a physical one-way valve function for this coordinate transformation, that is, each degradation channel can only make a positive contribution in the direction of exceeding the standard and cannot make a negative cancellation when it does not exceed the standard. This is consistent with the irreversible thermodynamic nature of the degradation process of ultrapure water film components.
[0043] Then, the system performs numerical validity checks and outputs the calculation results. The system verifies whether the deviation of reversible physical resistance is within the legal range of 0 to 45, and whether the deviation of irreversible chemical damage is within the legal range of 0 to 60. If an outlier occurs outside the legal range, a sensor drift warning is triggered, and the corresponding deviation is truncated to the range boundary. After successful verification, the system writes the deviations of reversible physical resistance and irreversible chemical damage into the intermediate calculation data structure of the current flaw detection cycle.
[0044] To facilitate understanding of the actual performance of the forward over-limit comparison operation under different membrane module operating conditions, specific values from this example are used for illustration: When the membrane module is in a healthy state during the initial commissioning phase, the typical value of the reversible fouling index is approximately 10 to 20, and the typical value of the irreversible aging index is approximately 5 to 15. After the forward over-limit comparison, both the reversible physical resistance deviation and the irreversible chemical damage deviation are linearly rectified to zero. At this point, the contribution of both degradation channels to system attenuation is zero. Subsequent contribution ratio calculations will yield a balanced result of 50% each due to the denominator approaching zero. However, since the deviation itself is zero, the subsequent decision tree will... The system is directly determined to be in a healthy state without triggering any intervention. When the membrane module is in the mid-to-late stage of operation and only reversible fouling occurs, the reversible fouling index of the module may reach 70 while the irreversible aging index remains at 20. At this time, the deviation of reversible physical resistance is 15 while the deviation of irreversible chemical damage is zero, and the reversible fouling channel will exclusively account for 100% of the attenuation contribution. When the membrane module experiences both fouling and aging simultaneously, the reversible fouling index is 65 and the irreversible aging index is 55. At this time, the deviation of reversible physical resistance is 10 and the deviation of irreversible chemical damage is 15. The two channels will contribute according to the relative magnitude of their respective deviations. The system directly transmits the deviations of reversible physical resistance and irreversible chemical damage to the subsequent two-dimensional attenuation contribution ratio quantification stage for processing.
[0045] Accordingly, in step S42, the deviation of reversible physical resistance and the deviation of irreversible chemical damage are quantified in two dimensions to obtain the contribution ratio of pollution attenuation and the contribution ratio of aging attenuation. It should be understood that although the deviation of reversible physical resistance and the deviation of irreversible chemical damage output in the previous step quantify the positive over-limit amplitude of the corresponding degradation channel beyond the safety boundary, the two deviations have unequal maximum value ranges due to their different critical thresholds. The maximum possible value of the deviation of reversible physical resistance is 45, while the maximum possible value of the deviation of irreversible chemical damage is 60. If the absolute deviation values are directly used for horizontal comparison between channels, a systematic judgment bias will occur due to the inconsistency of the measurement base, so that the aging channel with a larger measurement range naturally occupies a higher weight in terms of absolute value. Therefore, in the technical solution of this application, the deviation of reversible physical resistance and the deviation of irreversible chemical damage are further quantified in two dimensions to obtain the contribution ratio of pollution attenuation and the contribution ratio of aging attenuation. This converts the deviation values of the two paths with different absolute ranges into relative contribution shares expressed as percentages, so that the two degradation channels can achieve fair horizontal comparability in a normalized percentage space. In this way, the range bias effect caused by the difference in the setting of the critical threshold can be eliminated, so that the subsequent decision tree classification inference is based entirely on the relative responsibility ratio actually borne by each channel in the current system's overall excessive attenuation, rather than on the absolute deviation magnitude affected by the range.
[0046] More specifically, in a concrete example of this application, firstly, the current values of the reversible physical resistance deviation and the irreversible chemical damage deviation, received from the previous step, are read, and the two deviations are arithmetically summed to obtain the total absolute decay potential energy value generated by the current system exceeding the safety boundary. This total absolute decay potential energy value represents the sum of the over-limit increments after each of the two degradation channels exceeds the critical threshold within the current flaw detection cycle, and its physical meaning is the comprehensive severity of the membrane module's overall deviation from the normal service state. When performing the summation operation, the system simultaneously introduces a machine minimum constant as a zero-bias term added to the denominator; in this example, the value of this zero-bias term is 10. -6 Its function is to ensure that the division operation will not cause numerical overflow when both deviations are zero because the membrane module is in a completely healthy state.
[0047] Next, division is performed using the deviation of reversible physical resistance and the deviation of irreversible chemical damage as numerators, and the total absolute decay potential energy plus a zero-bias term as the denominator. The quotient is then multiplied by 100% to convert it into a percentage, thus obtaining the contribution ratios of pollution decay and aging decay. The mathematical expression of this two-dimensional decay contribution ratio quantification is as follows:
[0048] in, The calculated contribution percentage of pollution attenuation represents the relative percentage of responsibility borne by the reversible contamination pathway in the overall excessive attenuation of the current system, and its value ranges from 0% to 100%. The calculated contribution percentage of aging degradation represents the relative responsibility of the irreversible aging channel in the overall excessive degradation of the current system, and its value ranges from 0% to 100%. This refers to the deviation of the reversible physical resistance transmitted from the previous stage; This refers to the deviation of irreversible chemical damage transmitted from the previous step. To prevent the machine's minimum constant from having a denominator of zero, it is set to 10 in this example. -6 In this expression, the denominator part The absolute range differences of the two independent deviations are uniformly absorbed into the same normalized base, ensuring that regardless of the maximum possible deviation range of each channel, the sum of the proportions of the two channels after normalization is always equal to 100%, forming a complete and closed contribution allocation system. The numerator extracts the independent share of each channel's deviation in the normalized base, so that the channel with a larger deviation value receives a higher proportion, while the proportion of the channel with zero deviation is automatically reduced to zero. From the perspective of operation and maintenance decision-making for ultrapure water preparation systems, the core value of this normalization operation is that when the deviation of reversible physical resistance is 10 and the deviation of irreversible chemical damage is 15, although the deviation of the aging channel is higher than that of the fouling channel in absolute terms, after normalization, the contribution proportion of contamination attenuation is 40% and the contribution proportion of aging attenuation is 60%. This relative proportion truly reflects the actual responsibility weight of each channel for the overall performance degradation of the system at the current moment, providing dimensionless input for subsequent decision tree determination of dominant attribution.
[0049] Then, a completeness check is performed on the calculation results. The system verifies whether the sum of the contribution ratio of pollution attenuation and the contribution ratio of aging attenuation is equal to 100% within the allowable range of floating-point precision. If the verification fails, it indicates that there is a data truncation or overflow anomaly in the calculation link. The system will trigger a computational integrity alarm and roll back to the historical results of the previous flaw detection cycle as a conservative alternative. At the same time, the system verifies whether the output values of the two proportions are correctly processed to the balanced default value of 50% when both the deviation of reversible physical resistance and the deviation of irreversible chemical damage are zero. This default value will not trigger any single-channel dominant judgment in the subsequent decision tree because 50% is lower than the preset absolute dominant boundary of 70%. The system will determine it as a healthy state or a dual-track mixed decline state and will not execute aggressive unidirectional intervention measures.
[0050] To facilitate understanding of the ability of this contribution ratio quantification to distinguish between different operating states, typical operating data from this example is used for illustration: When the membrane module only experiences reversible fouling while the substrate remains intact, the module's reversible fouling index reaches 70, corresponding to a reversible physical resistance deviation of 15, while the module's irreversible aging index is 20, corresponding to an irreversible chemical damage deviation of 0. At this point, the contribution ratio of fouling attenuation is 100%, and the contribution ratio of aging attenuation is 0%, clearly indicating that the reversible fouling channel bears all the responsibility for attenuation. When the membrane module simultaneously experiences both fouling and aging degradation, the reversible physical resistance deviation is 10, and the irreversible chemical damage deviation is 15. At this point, the contribution ratio of fouling attenuation is 40%, and the contribution ratio of aging attenuation is 60%, clearly reflecting that the aging channel bears a higher proportion of attenuation responsibility but has not yet reached absolute dominance. The system writes the contribution ratios of fouling attenuation and aging attenuation into the intermediate calculation data structure of the current flaw detection cycle, directly passing them to the subsequent operation and maintenance strategy classification decision tree inference stage for processing.
[0051] Accordingly, in step S43, based on the operation and maintenance strategy classification decision tree model, the contribution ratios of pollution attenuation and aging attenuation are compared with the preset absolute dominant boundary, and enumeration classification inference is completed to obtain the dominant membrane state definition result. It should be understood that although the contribution ratios of pollution attenuation and aging attenuation output in the previous step have quantified the relative responsibility ratios of the two degradation channels into a percentage space, the percentage values themselves are still real values in the continuous variable domain. The operation and maintenance control system cannot directly select discrete strategy execution branches based on continuous values, such as chemical cleaning sequence activation, physical locking of the cleaning system and component replacement early warning, or a comprehensive conservative strategy, because the selection of strategy branches is essentially a classification mapping problem from a continuous numerical space to a finite discrete enumeration set. Therefore, in the technical solution of this application, the contribution ratios of pollution attenuation and aging attenuation are further compared with the preset absolute dominant boundary based on the operation and maintenance strategy classification decision tree model, and enumeration classification inference is completed to obtain the dominant membrane state definition result. This establishes a deterministic interval division boundary in the percentage continuous space, mapping the intervals into which the two proportion values fall to discrete classification labels with clear operation and maintenance strategy directions. In this way, the contribution ratio information of the continuous domain can be compressed into an enumerated type judgment conclusion that can be directly consumed in the subsequent closed-loop control instruction generation steps, making the selection of strategy branches deterministic and traceable.
[0052] More specifically, in a concrete example of this application, firstly, a preset absolute dominance boundary is loaded from the configuration parameter table of the local knowledge base. The engineering basis for setting this preset absolute dominance boundary is: when the contribution ratio of a single degradation channel exceeds this boundary value, the channel can be identified as the absolute dominant factor in the current membrane module performance degradation with engineering confidence. The corresponding operation and maintenance measures can and should be designed specifically around the characteristics of this single channel without considering the influence of the other channel. In this example, the preset absolute dominance boundary is set to 70%. This value is selected based on the following engineering considerations: in the actual operation and maintenance scenario of membrane module degradation, when the contribution ratio of a certain channel exceeds 70%, it means that the contribution ratio of another channel is less than 30%. At this time, the impact of the secondary channel on the overall system degradation has been reduced to a negligible level. The comprehensive benefits brought by implementing unidirectional intervention measures for the dominant channel are far higher than adopting a compromise two-way conservative strategy.
[0053] Next, based on the logical structure of the operation and maintenance strategy classification decision tree model, multi-path conditional judgment and enumeration classification inference are performed on the contribution ratios of contamination attenuation and aging attenuation. The first-level decision node of the decision tree reads the value of the contribution ratio of contamination attenuation and compares it with the preset absolute dominance boundary of 70%. If the contribution ratio of contamination attenuation is greater than 70%, it directly reaches the leaf node along the left subtree path and outputs the enumeration classification label as reversible fouling dominance, corresponding to machine state code 0x01. In the operation and maintenance semantics of the ultrapure water preparation system, this judgment result is characterized as the accumulation of reversible surface deposits on the membrane surface being the absolute main cause of the current capacity decline, and in-situ chemical cleaning should be performed to restore flux. If the first-level judgment fails, the decision tree enters the second-level judgment node to read the value of the contribution ratio of aging degradation and compares it with the preset absolute dominance boundary of 70%. If the contribution ratio of aging degradation is greater than 70%, it reaches the leaf node along the left subtree path and outputs the enumeration classification label "Irreversible Aging Dominance," corresponding to machine state code 0x02. This judgment result indicates that the irreversible loss of desalination capability of the membrane substrate due to chemical degradation or mechanical damage is the absolute main cause of the current capacity reduction. Subsequently, the chemical cleaning system should be shut down to prevent ineffective cleaning damage and a component replacement warning should be initiated. If the second-level judgment also fails, that is, the contribution ratios of contamination degradation and aging degradation are both less than 70% and are in the middle region between 30% and 70%, the decision tree reaches the third leaf node along the default path and outputs the enumeration classification label "Dual-track Hybrid Degradation," corresponding to machine state code 0x03. This judgment result indicates that both degradation channels have made significant contributions to the current performance degradation, but neither has an absolute dominant position. Subsequently, a comprehensive conservative strategy that takes into account cleaning maintenance and life management should be adopted.
[0054] Then, the enumeration classification code inferred from the decision tree, along with the specific values of the pollution attenuation contribution ratio and the aging attenuation contribution ratio that triggered the determination, are packaged as confidence attributes to generate a structured dominant membrane state delineation result data packet. The data packet's field structure includes: an enumeration classification code field (1 byte storing one of the three valid states: 0x01, 0x02, or 0x03); a pollution contribution ratio field (4 bytes storing a single-precision floating-point number); an aging contribution ratio field (4 bytes storing a single-precision floating-point number); and an 8-byte timestamp field for the current flaw detection cycle. The system performs a validity check on the packaged dominant membrane state delineation result, verifying whether the enumeration code belongs to one of the three predefined valid states and whether the sum of the two ratio values is equal to 100% within the floating-point precision range. After passing the verification, the result is written to the current cycle record in the operating status database.
[0055] Specifically, in the membrane module status monitoring scenario of an ultrapure water treatment device, the first embodiment of sub-step S43 in the original scheme above adopts a decision tree model with a fixed absolute dominance boundary (e.g., a single contribution exceeding 70%) to classify and determine the contribution ratios of fouling decay and aging decay through independent channels. Its output only includes three enumerated states: reversible fouling-dominated, irreversible aging-dominated, and dual-track mixed decay. This mechanism, at its underlying logic, treats the reversible fouling channel and the irreversible aging channel as orthogonal and independent degradation axes, determining the dominant decay attribution of the current membrane module by individually comparing their respective proportions with a fixed threshold.
[0056] However, in actual operation of reverse osmosis membranes and continuous electrostatic desalination resins, a nonlinear positive feedback coupling acceleration relationship, confirmed in the field of membrane science, exists between reversible fouling and irreversible aging. Specifically, when organic colloids or microorganisms form an adhesion layer on the polyamide membrane surface, their anaerobic metabolites and the acidic or oxidizing microenvironment induced by local concentration polarization on the membrane surface, even when macroscopic residual chlorine monitoring results are zero, continue to launch hydrolytic attacks on the polymer cross-linking bonds of the membrane substrate, thus significantly accelerating the irreversible aging process. This constitutes a positive feedback path of fouling promoting aging. Conversely, once the membrane surface suffers microscopic roughness damage and pore size expansion due to initial chemical degradation or water hammer impact, the uneven nucleation sites formed on the rough surface provide adhesion conditions far superior to those of a healthy, smooth membrane surface for subsequent colloidal deposition and microbial anchoring. This results in a rate of secondary fouling and biofilm growth that far exceeds the natural accumulation rate of simple fouling channels, constituting a positive feedback path of aging promoting fouling.
[0057] In other words, the fundamental weakness of the first embodiment lies in the fact that when both deviations are at a moderate level of activity (e.g., approximately 50% each), neither crosses the fixed dominant boundary of 70%, and the original decision tree can only classify it as a fuzzy dual-track mixed degradation. However, in the real physicochemical scenario of the membrane micro-region, this coexistence state is precisely the region where the positive feedback coupling effect is most active. Since the two degradation modes promote and accelerate each other in the same micro-region of the membrane, their combined deterioration rate may far exceed the linear cumulative rate when any single channel is dominant. The first embodiment completely lacks the ability to perceive the product-type interaction term between the two deviations in its mathematical structure. It cannot quantify the strength of the coupling acceleration effect, nor can it dynamically adjust the judgment sensitivity according to the coupling strength. Ultimately, this results in the system being completely blind to the hidden danger of the membrane module's impending sudden failure due to positive feedback deterioration.
[0058] To address the aforementioned shortcomings, a second embodiment is proposed. More specifically, in the second embodiment of this application, Figure 5This is a flowchart illustrating the operation and maintenance strategy classification decision tree model of the intelligent monitoring method for the operation status of an ultrapure water treatment device according to an embodiment of this application. It compares and determines the contribution ratios of pollution attenuation and aging attenuation with a preset absolute dominant boundary, and completes enumeration classification inference to obtain the dominant membrane state definition result. Figure 5 As shown, step S43 includes: S431, calculating the coupling synergy strength of the two degradation channels for the deviation of reversible physical resistance and the deviation of irreversible chemical damage to obtain the pollution-aging coupling synergy factor; S432, based on the pollution-aging coupling synergy factor, correcting the proportion of pollution attenuation contribution and the proportion of aging attenuation contribution by coupling gain and compressing the basic dominant boundary to obtain the coupled corrected pollution contribution proportion, the coupled corrected aging contribution proportion, and the adaptive dominant determination boundary; S433, performing multi-path condition comparison and enumeration classification inference on the coupled corrected pollution contribution proportion, the coupled corrected aging contribution proportion, the adaptive dominant determination boundary, and the pollution-aging coupling synergy factor to obtain the dominant membrane state definition result.
[0059] In step S431, the coupling synergy strength of the two degradation channels—reversible physical resistance deviation and irreversible chemical damage deviation—is calculated to obtain the fouling-aging coupling synergy factor. It should be understood that in the aforementioned first embodiment, when the operation and maintenance strategy classification decision tree performs independent channel classification judgments on the contribution ratio of fouling attenuation and the contribution ratio of aging attenuation, its underlying mathematical structure treats the reversible fouling channel and the irreversible aging channel as orthogonal and independent degradation coordinate axes, completely lacking the ability to perceive the product-type interaction term between the two deviations. However, in the actual degradation physical process of reverse osmosis membrane modules, there is a nonlinear positive feedback coupling acceleration relationship between the two degradation mechanisms of reversible fouling and irreversible aging, which has been confirmed in the field of membrane science. Therefore, in the technical solution of this application, the coupling synergy strength of the two degradation channels—reversible physical resistance deviation and irreversible chemical damage deviation—is calculated to obtain the fouling-aging coupling synergy factor. This extracts new dimensional information characterizing the coupling interaction strength from the original two independent deviation signals, providing a physical quantitative basis for subsequent boundary modulation and judgment correction. In this way, without adding any physical sensors, the activity level of the positive feedback coupling loop can be captured from the product-type interaction relationship of the two deviations by using software calculation dimensions, enabling the monitoring system to perceive the state of implicit coupling accelerating deterioration.
[0060] From the physicochemical essence of membrane separation engineering, the local microenvironment only possesses the thermodynamic conditions to activate a positive feedback coupling loop when both degenerate channels are simultaneously active within the same microregion on the membrane surface. Therefore, the quantification of coupling interaction strength cannot simply rely on the linear summation of the two deviations; instead, a mathematical operator must be chosen that naturally returns to zero when any one path is zero. Based on this physical constraint, a saturated exponential geometric mean interaction kernel function is used to calculate the coupling synergy strength for the deviations of reversible physical resistance and irreversible chemical damage. The geometric mean operator naturally satisfies the condition that the interaction term disappears when either path is silent, while the outer saturation exponent ensures that the factor does not diverge unbounded as the deviations of the two paths continue to increase, but rather asymptotically approaches the physical upper limit.
[0061] In a specific example of this application, the implementation process of this step is as follows: First, read the current values of the reversible physical resistance deviation and the irreversible chemical damage deviation passed in from the previous step. Multiply the two deviations and take the arithmetic square root of the product to calculate the geometric mean of the two deviations. The physical meaning of this geometric mean calculation is as follows: When the reversible physical resistance deviation is zero, that is, when the reversible fouling of the membrane surface has not exceeded the critical cleaning threshold, no matter how large the value of the irreversible chemical damage deviation is, the product is always zero, resulting in a geometric mean of zero. At this time, the fouling channel fails to provide the catalytic conditions of the membrane surface microenvironment for the aging channel, the entrance channel of the positive feedback coupling loop is cut off, and the factor should disappear naturally; symmetrically, the same applies when the irreversible chemical damage deviation is zero. Only when both deviations simultaneously obtain non-zero positive values will the geometric mean produce a non-zero output, and its value increases rapidly with the synchronous increase of the two deviations, accurately mapping the nonlinear amplification characteristics of the coupling effect under the dual-channel coexistence state.
[0062] Next, the geometric mean result is multiplied by the coupling sensitivity gain constant, and the negative value is taken as the exponent of the natural exponential function. Then, the saturation mapping operation is completed by subtracting this exponential function value from 1. The process is as follows:
[0063] in, The calculated pollution-aging coupling synergy factor has a value range of [value range missing]. The closer the value is to 1, the more intense the positive feedback coupling effect on the membrane surface; a value of 0 indicates that there is no interactive coupling between the two degenerate channels. The deviation of the reversible physical resistance input from the upstream sub-step; The deviation of irreversible chemical damage introduced by the upstream sub-step; The coupling sensitivity gain constant reflects the steepness of the response of a specific membrane substrate to the coexistence of fouling and aging, and is calibrated using accelerated life test data from the membrane module at the time of manufacture. It is worth noting that this saturated exponential mapping structure ensures that the output value is strictly constrained within a closed interval of 0 to 1: when both deviations are zero or only one is zero, the exponent term is zero. The output value is 1-1=0; when the deviation of the two paths increases simultaneously, causing the geometric mean to continue to rise, the exponential term increases negatively. Approaching zero, the output value asymptotically approaches 1 but never exceeds 1. This upper bound saturation characteristic corresponds, at the physical level, to a thermodynamic limit for positive feedback coupling effects; even with extremely severe film degradation, the coupling acceleration efficiency cannot exceed the theoretical upper limit of chemical reaction kinetics. In this expression, Part of it is the geometric mean of the two deviations. The operator satisfies the physical constraint that if either path is zero, the whole is zero. At the same time, when the two deviations are equal, the geometric mean reaches the maximum value under a given total value. This reflects the experimental observation law that the coupling effect is most intense when the two degradation mechanisms coexist in equilibrium in the same micro-region of the membrane surface in the field of membrane separation. The linear scaling factor, acting as the geometric mean, controls the steepness of the rise in the saturation curve. A higher value indicates a rapid increase in the factor even at lower deviation levels, corresponding to membrane substrate types that are highly sensitive to the contamination-aging coupling phenomenon. In this example, this value was calibrated to 0.15 through accelerated life testing; outer layer The saturation exponential structure provides a monotonically increasing bounded mapping from 0 to 1 in the low-coupling region. Time approximately linear growth Provides good detection sensitivity in the high coupling range The physical upper limit of 1, which asymptotically saturates at 1, avoids numerical instability caused by the exponential amplification of deviation measurement noise in the high-value region.
[0064] Then, the calculated pollution-aging coupling synergy factor is numerically validated. The system verifies whether the factor is strictly within a closed interval of 0 to 1. If a small negative value or a value greater than 1 occurs due to floating-point arithmetic precision, it is truncated to the nearest boundary point. The system also records the slope of the factor's change trend over the most recent consecutive flaw detection cycles. If the slope remains positive, it indicates that the coupling effect is intensifying, providing additional dynamic trend reference for subsequent boundary compression amplitude modulation.
[0065] To facilitate understanding of the actual response behavior of this coupling synergy factor under different membrane module operating conditions, typical operating data from this example are used for illustration: when the membrane module only experiences reversible fouling while the substrate remains intact, the reversible physical resistance deviation is 15 while the irreversible chemical damage deviation is 0. The pollution-aging coupling synergy factor is 0, which correctly reflects the physical fact that single-channel degradation does not trigger a coupling effect; when both deviations are at a moderate activity level, such as a reversible physical resistance deviation of 10 and an irreversible chemical damage deviation of 15, Substituting into the formula, we get This high value indicates that the positive feedback coupling loop on the membrane surface is already highly active. Therefore, the judgment boundary should be significantly compressed to improve the system's response sensitivity to this latent hazard. Through the above calculations, without adding any physical sensors, software calculations can extract new dimensional information characterizing the coupling interaction strength from the original two independent deviation signals, providing a physical quantitative basis for subsequent boundary modulation and judgment correction. The system writes the contamination aging coupling synergy factor into the intermediate computational data structure of the current flaw detection cycle, directly transmitting it to the subsequent coupling sensing contribution ratio correction and adaptive boundary compression stages for processing.
[0066] In step S432, the pollution attenuation contribution ratio and aging attenuation contribution ratio are corrected by coupling gain based on the pollution-aging coupling synergy factor, and the basic dominance boundary is compressed to obtain the coupled corrected pollution contribution ratio, the coupled corrected aging contribution ratio, and the adaptive dominance judgment boundary. It should be understood that, under high coupling conditions, the original pollution attenuation contribution ratio and aging attenuation contribution ratio are diluted because the denominator includes the other's value, and their absolute values cannot truly reflect the actual hazard level amplified by positive feedback coupling. Specifically, when two degradation channels are simultaneously active and the positive feedback coupling loop is established, the actual hazard level of each channel being accelerated by the other's presence is far higher than the level represented by its independent proportion value. However, the mathematical structure of the normalization operation forces the sum of the two proportions to be constrained to 100%, causing their absolute values to be suppressed by the coexistence contribution of the other. At the same time, when the positive feedback coupling loop is truly established, the dynamics of membrane degradation are far more critical than when a single channel dominates; maintaining the original fixed judgment threshold will cause the system to lose the time window for early intervention. Therefore, in the technical solution of this application, the contribution ratios of pollution attenuation and aging attenuation are further corrected by the coupling gain based on the pollution-aging coupling synergy factor, and the basic dominant boundary is compressed to obtain the coupled corrected pollution contribution ratio, the coupled corrected aging contribution ratio, and the adaptive dominant judgment boundary. This allows for nonlinear upward adjustment of the two proportions using the quantified coupling synergy strength information to restore the true hazard level after positive feedback amplification. Simultaneously, the absolute dominant judgment boundary is dynamically compressed according to the strength of the coupling factor to improve the system's response sensitivity to the coupled accelerated deterioration state. This ensures that the corrected contribution ratio values truly reflect the actual attenuation severity of each channel under the coupling amplification effect, while simultaneously shifting the judgment boundary forward synchronously with the coupling strength, moving the warning threshold forward to before the inflection point of accelerated degradation, and significantly shortening the decision delay from latent deterioration to maintenance intervention.
[0067] In a specific example of this application, the implementation process of this step is as follows: First, the pollution aging coupling synergy factor passed in from the previous step, as well as the pollution attenuation contribution ratio and aging attenuation contribution ratio calculated in the preceding step, are read. Simultaneously, two coupling gain response weight parameters and boundary compression coefficient parameters are loaded from the configuration parameter table of the local knowledge base. The system performs a coupling gain correction operation on the pollution attenuation contribution ratio. The correction algorithm uses a structure of coupling gain multiplied by the remaining growth space to ensure that the corrected value never exceeds the upper bound of the full scale of the physically reasonable domain. Specifically, the pollution aging coupling synergy factor is multiplied by the coupling gain response weight of the reversible contamination channel to obtain the coupling gain amplitude of the channel. Then, this gain amplitude is multiplied by the remaining growth space of the pollution attenuation contribution ratio from 100% of the full scale to obtain the actual correction increment of the channel. Finally, this correction increment is superimposed on the original pollution attenuation contribution ratio to generate the coupled corrected pollution contribution ratio. The system performs the same correction operation on the aging attenuation contribution ratio to generate the coupled corrected aging contribution ratio. The mathematical expression of the above coupling gain correction is:
[0068] in, To couple and correct the proportion of pollution contribution, characterizing the actual attenuation and hazard level of the reversible clogging channel after reduction by the positive feedback coupling amplification effect; The proportion of aging contribution is used to correct the coupling effect, which characterizes the actual attenuation and damage level of the irreversible aging channel after being restored by the positive feedback coupling amplification effect. The percentage of the original pollution attenuation; The percentage of the original aging degradation contribution; This refers to the pollution-aging coupling synergy factor obtained from the previous step. The coupling gain response weight of the reversible fouling channel corresponds to the engineering strength coefficient of the positive feedback path of accelerated secondary fouling by the aging rough surface. In this example, it is calibrated to 0.6 based on the experimental data on the correlation between film surface roughness and colloidal deposition rate. The coupling gain response weight of the irreversible aging channel corresponds to the engineering strength coefficient of the positive feedback path of the fouling microenvironment accelerating the degradation of the membrane substrate. In this example, it is calibrated to 0.5 based on the experimental data of the promotion of the hydrolysis rate of polyamide by the anaerobic metabolites of the membrane biofilm. These represent the remaining growth potential of each channel's current percentage from full scale, serving as a natural cutoff for gain correction. In this expression, The overall strength of the coupling synergy factor is allocated through channel-specific weights to obtain the upward adjustment gain coefficient that the reversible fouling channel should obtain due to the coupling effect. The physical meaning of this gain coefficient is: when there is aging damage on the film surface at the same time, the uneven nucleation sites of the aging rough surface provide convenient conditions for colloidal deposition, and the actual accumulation rate of reversible fouling is amplified by how many times compared to independent channels. The multiplicative truncation, partially acting as a gain, ensures a crucial physical constraint: when the original contamination attenuation contribution approaches 100%, the remaining growth space approaches zero. Regardless of the magnitude of the coupling gain's correction increment, it naturally approaches zero, thus guaranteeing that the coupling correction's contamination contribution will never exceed the upper bound of the physically reasonable domain of 100%. Symmetrically, The same logic is applied to the irreversible aging channel. The physical meaning of this modification is that when reversible fouling exists on the membrane surface at the same time, the accelerated hydrolytic attack effect of the local acidic oxidative microenvironment formed by anaerobic metabolites below the fouling layer on the polymer crosslinking bonds of the membrane substrate is quantified and reduced to the contribution ratio value. and The different values reflect the asymmetry of the two positive feedback paths in terms of engineering strength. In the polyamide composite film system of this example, the promoting effect of aging roughness on secondary fouling is slightly stronger than the promoting effect of fouling microenvironment on substrate degradation.
[0069] Next, while correcting the contribution ratio, a dynamic compression calculation of the basic dominance boundary is performed. The system reads the system-preset fixed basic absolute dominance boundary and boundary compression coefficient from the configuration parameter table. The absolute dominance judgment boundary needs to be dynamically compressed according to the strength of the coupling factor. The physical consideration is that when the positive feedback coupling loop is truly established, the overall deterioration rate of membrane degradation far exceeds that of the single-channel dominance case. If the original fixed judgment threshold of 70% is maintained at this time, the system will only give a classification judgment after the membrane degradation has entered an accelerated deterioration trajectory, losing the time window for early intervention. The system squares the pollution-aging coupling synergy factor, multiplies it by the boundary compression coefficient to obtain the compression ratio, then subtracts this compression ratio from 1 and multiplies it by the fixed basic absolute dominance boundary to obtain the adaptive dominance judgment boundary. The mathematical expression of this dynamic boundary compression is:
[0070] in, The adaptive dominant decision boundary after dynamic compression is a value that decreases as the coupling strength increases, causing the decision sensitivity to shift forward synchronously with the coupling effect. The fixed foundation absolute dominant boundary is preset for the system (such as 70% in the original scheme); The boundary compressibility coefficient has a value range of 1. The maximum modulation amplitude of the coupling effect on the determination sensitivity is controlled. In this example, the value is 0.4, which means that when the coupling factor reaches the full scale of 1, the boundary is compressed to 60% of the original value, i.e., 42%. The physical consideration for choosing quadratic over linear coupling factors is that the system remains controllable in the weakly coupled region, and premature boundary compression should be avoided to prevent false triggering. At that time, its square was only 0.09, resulting in only a slight compression of about 2.5 percentage points for 70% of the basic boundary. The system's judgment behavior was basically consistent with the first embodiment, avoiding unnecessary emergency responses triggered by excessive boundary sensitization when the coupling effect was not yet significant; however, once the coupling factor entered the strong effect region, such as When its square is 0.64, the boundary is significantly reduced to 70%×(1-0.4×0.64)=52.1%. At this point, as long as the correction percentage of any channel exceeds 52.1%, the dominant judgment will be triggered, thus moving the warning threshold forward to before the inflection point of accelerated degradation. The engineering significance of the parameter is to set a safety upper limit on the boundary compression magnitude, even under extreme coupling conditions. The time boundary will not be compressed to zero but will be preserved. The minimum threshold value is set to prevent the system from frequently triggering the dominant decision under sensor noise interference. The multiplication structure ensures the monotonically decreasing characteristic of the compression operation. The larger the coupling factor, the smaller the multiplier, the lower the boundary, and the more sensitive the system, forming an adaptive linkage mechanism with a positive correlation between coupling strength and decision sensitivity.
[0071] Then, a rationality check is performed on the three calculated output data. The system verifies whether the proportion of pollution contribution from the coupled correction is within a legal range between the original pollution attenuation contribution and 100%, and whether the proportion of aging contribution from the coupled correction is within a legal range between the original aging attenuation contribution and 100%. The monotonic gain characteristic of the correction operation ensures that the proportion after correction is necessarily greater than or equal to the proportion before correction. The system also verifies whether the adaptive dominant decision boundary is within... and Within the legal range, in this example, it is between 42% and 70%.
[0072] To facilitate understanding of the combined effect of the coupling correction and boundary compression mechanism, typical operating data from this example are used: When the pollution-aging coupling synergy factor is 0.84 and the original pollution attenuation contribution is 40% and the original aging attenuation contribution is 60%, the coupling correction pollution contribution is 40% + 0.84 × 0.6 × (1 - 40%) = 70.24%, the coupling correction aging contribution is 60% + 0.84 × 0.5 × (1 - 60%) = 76.8%, and the adaptive dominant decision boundary is 70% × (1 - 0.4 × 0.84). 2 =50.24%. In this numerical scenario, the original scheme could only be judged as dual-track hybrid decay because the proportions of both paths did not exceed 70%. However, after coupling correction, the proportions of both paths exceeded the compressed 50.24% boundary. The subsequent four-state classification inference will be able to identify this hidden danger state of accelerated deterioration due to coupling. The system writes the coupling-corrected pollution contribution proportion, the coupling-corrected aging contribution proportion, and the adaptive dominant judgment boundary into the intermediate operation data structure of the current flaw detection cycle, and passes them together with the pollution-aging coupling synergy factor from the previous stage to the subsequent multi-path condition comparison and enumeration classification inference stage for processing.
[0073] In step S433, the proportion of pollution contribution due to coupling correction, the proportion of aging contribution due to coupling correction, the adaptive dominance judgment boundary, and the pollution-aging coupling synergy factor are compared under multiple conditions and enumerated and classified to obtain the dominant membrane state definition result. It should be understood that since the contribution proportion correction and boundary adaptive modulation have been completed in the previous step, the three data of the proportion of pollution contribution due to coupling correction, the proportion of aging contribution due to coupling correction, and the adaptive dominance judgment boundary have been adjusted from the original proportion and fixed boundary to new values that can reflect the positive feedback coupling amplification effect. However, the above data are still real variables in the continuous domain and have not yet been transformed into discrete classification labels that can directly drive the branch selection of differentiated operation and maintenance strategies. Moreover, compared with the first embodiment, which only includes the classification granularity of three enumerated states: reversible fouling dominance, irreversible aging dominance, and dual-track hybrid decay, the second embodiment needs to introduce a fourth new enumeration category that can identify the hidden danger state that the positive feedback coupling loop has been established, so that the subsequent closed-loop control instruction set can generate a comprehensive emergency strategy that is different from single-channel dominance for this special state. Therefore, in the technical solution of this application, multi-path condition comparison and enumeration classification inference are further performed on the proportion of coupling-corrected contamination contribution, the proportion of coupling-corrected aging contribution, the adaptive dominant judgment boundary, and the contamination-aging coupling synergy factor to obtain the dominant membrane state definition result. This completes the deterministic classification mapping from continuous numerical space to a four-state discrete enumeration set. Moreover, the classification inference process prioritizes the evaluation of the coupling-accelerated deterioration state to ensure that the state with the highest danger level is not obscured by conventional branch logic. In this way, the condition monitoring system of the ultrapure water treatment device can perceive the contamination-aging positive feedback coupling effect, which was originally ignored by the first embodiment, in the pure software calculation dimension. This triggers accurate early warning when the membrane degradation is still in the early inflection point stage of coupling acceleration, rather than being passively informed by the vague dual-track hybrid degradation label after the component capacity has suffered an irreversible precipitous collapse.
[0074] More specifically, in a specific example of this application, the implementation process of this step is as follows: First, the four data points passed from the previous step—the proportion of pollution contribution due to coupling correction, the proportion of aging contribution due to coupling correction, the adaptive dominant decision boundary, and the pollution-aging coupling synergy factor—are loaded into the input register of the extended multi-path conditional decision logic. Simultaneously, the system loads the coupling critical threshold from the configuration parameter table of the local knowledge base. The physical meaning of this coupling critical threshold is that when the pollution-aging coupling synergy factor exceeds this value, the membrane surface positive feedback coupling loop can be considered to be in an active establishment state with engineering confidence. In this example, this threshold is set to 0.6. The basis for selecting this value is: when the coupling synergy factor reaches 0.6, according to the saturation exponential mapping model of the previous step, the geometric mean of the two deviations has reached... The level corresponds to a severe coexistence state where both degradation channels have penetrated beyond their respective critical thresholds.
[0075] Next, the classification inference process prioritizes evaluating the accelerated deterioration state. The system first compares the contamination-aging coupling synergy factor with the coupling crisis threshold of 0.6 to determine if the positive feedback coupling loop is active. If this condition is met, the system further simultaneously checks whether the proportion of contamination contribution from coupling correction exceeds the adaptive dominance boundary and whether the proportion of aging contribution from coupling correction exceeds the adaptive dominance boundary. When all three conditions are met—that is, the contamination-aging coupling synergy factor exceeds the coupling crisis threshold and both the proportions of contamination contribution from coupling correction and aging correction exceed the adaptive dominance boundary—the system determines the current membrane state to be accelerated deterioration, corresponding to the enumeration classification code 0x04. In the ultrapure water preparation scenario, this state signifies that the membrane surface positive feedback loop has been established and is on an exponential deterioration trajectory. Specifically, the anaerobic metabolites of the organic colloidal deposits on the membrane surface are continuously corroding the underlying polyamide substrate. Simultaneously, the roughened surface formed by corrosion accelerates the adsorption of new colloidal deposits. The two positive feedback paths feed each other, forming a self-accelerating vicious cycle, whose overall deterioration rate far exceeds the linear cumulative rate when any single channel is dominant. The corresponding maintenance response needs to be upgraded from a single cleaning or a single spare part to a comprehensive emergency measure that simultaneously performs in-situ cleaning, immediately reduces transmembrane pressure differential, and initiates spare part procurement for membrane modules in advance. In-situ cleaning is used to weaken the reversible fouling layer as soon as possible to cut off the positive feedback loop of pollution promoting aging. Reducing transmembrane pressure differential is used to slow down the concentration polarization intensity of the membrane surface to suppress the attack rate of the local acidic microenvironment on the substrate. Initiating spare part procurement in advance is based on the prediction that the remaining service life of the membrane module will be significantly shortened under this accelerated deterioration trajectory.
[0076] Then, if the triple conditions for the above-mentioned coupling-accelerated deterioration state are not triggered simultaneously, the system degenerates into regular branch decision logic. The system compares the contribution ratio of coupled-corrected contamination with the adaptive dominance threshold. If the contribution ratio of coupled-corrected contamination alone exceeds the adaptive dominance threshold, the current membrane state is determined to be dominated by reversible fouling, corresponding to enumeration classification code 0x01. This determination indicates that the accumulation of reversible surface deposits is still the main cause of current capacity reduction, and a targeted chemical cleaning sequence should be executed subsequently. If the contribution ratio of coupled-corrected contamination does not exceed the threshold, the system further compares the contribution ratio of coupled-corrected aging with the adaptive dominance threshold. If the contribution ratio of coupled-corrected aging alone exceeds the adaptive dominance threshold, the current membrane state is determined to be dominated by irreversible aging, corresponding to enumeration classification code 0x02. This determination indicates that chemical degradation of the membrane substrate is the main cause, and the cleaning system should be locked down and a replacement warning should be initiated. If neither exceeds the adaptive dominance threshold, the current membrane state is determined to be a dual-track mixed decay, corresponding to enumeration classification code 0x03. This determination indicates that both channels contribute to a certain extent but neither has reached a dominant position and the coupling effect is not yet significant, and a conservative comprehensive maintenance strategy should be adopted subsequently.
[0077] Furthermore, the four enumerated classification labels derived from the decision inference, along with the values of the pollution-aging coupling synergy factor, are encapsulated as confidence-added attributes to generate a structured dominant membrane state delineation result data packet. The field structure of this data packet is expanded from the first embodiment as follows: an enumerated classification code field (1 byte storing one of the four legal states from 0x01 to 0x04), a coupling-corrected pollution contribution ratio field (4 bytes), a coupling-corrected aging contribution ratio field (4 bytes), a pollution-aging coupling synergy factor field (4 bytes), an adaptive dominant judgment boundary field (4 bytes), and a current flaw detection cycle timestamp field (8 bytes). The system performs a legality check on the encapsulated dominant membrane state delineation result, verifying whether the enumerated code belongs to one of the four predefined legal states. If the check passes, the result is written to the current cycle record in the operating status database.
[0078] Through the aforementioned four-state coupled sensing classification and inference, the adaptive boundary compression mechanism enables the judgment sensitivity to advance synchronously with the coupling strength, significantly shortening the decision delay from latent deterioration to operation and maintenance intervention. This effectively avoids secondary chemical cleaning damage to the aged membrane surface caused by ineffective chemical cleaning due to classification blind spots, and prevents membrane modules from operating beyond their service life on unidentified accelerated deterioration tracks until sudden breakthrough failure. The introduction of four-state classification granularity allows the subsequent closed-loop control instruction set to generate comprehensive emergency strategies that differ from those dominated by single channels for coupled accelerated deterioration states, thereby improving the overall accuracy of operation and maintenance decisions for the ultrapure water preparation system and the economic utilization rate of the membrane module throughout its entire life cycle. The system transmits the dominant membrane state definition results, along with the module's reversible fouling index and irreversible aging index for this flaw detection cycle, to the subsequent branch strategy optimization steps for consumption.
[0079] In summary, through the second embodiment described above, the condition monitoring system of the ultrapure water treatment unit can perceive the previously ignored positive feedback coupling effect of fouling and aging at the purely software computational level. This allows for precise early warnings triggered when membrane degradation is still in the early inflection point of accelerated coupling, rather than passively being notified by the fuzzy hybrid degradation labels in the original solution after an irreversible precipitous collapse in module capacity. In terms of engineering effectiveness, the adaptive boundary compression mechanism shifts the judgment sensitivity forward in tandem with the coupling strength, significantly shortening the decision delay from latent deterioration to maintenance intervention. This effectively avoids secondary chemical damage to aged membrane surfaces caused by ineffective chemical cleaning due to blind spots in classification, and prevents membrane modules from operating beyond their designed lifespan on unidentified accelerated deterioration paths until sudden breakthrough failure. The introduction of four-state classification granularity enables the subsequent closed-loop control instruction set to generate comprehensive emergency strategies different from those dominated by single channels for accelerated coupling deterioration states, thus improving the overall accuracy of maintenance decisions for the ultrapure water preparation system and the economic utilization rate of the membrane modules throughout their entire lifecycle.
[0080] Specifically, in step S5, a branching strategy optimization is performed on the dominant membrane state definition result, the component reversible contamination index, and the component irreversible aging index to generate an adaptive closed-loop control instruction set. The adaptive closed-loop control instruction set is then sent to the programmable logic controller (PLC) for execution. It should be understood that although the dominant membrane state definition result output in the previous step has assigned the current degradation state of the membrane module to a specific enumeration classification label, this classification label itself is only a diagnostic conclusion at the information level and has not yet been converted into engineering control parameters that can directly drive the underlying physical actuators. The PLC cannot accept the enumeration label as execution input; what it requires are specific frequency setpoints, dosing pump start / stop signals, valve opening / closing sequences, and durations—deterministic numerical instructions that can be directly mapped to hardware register addresses. Therefore, in the technical solution of this application, a branching strategy optimization is further performed on the dominant membrane state definition results, the reversible pollution index of the module, and the irreversible aging index of the module to generate an adaptive closed-loop control instruction set. This adaptive closed-loop control instruction set is then sent to the programmable logic controller for execution. This transforms the classification judgment conclusions at the diagnostic level into a comprehensive control instruction package, including chemical cleaning parameters, hardware interlock signals, and frequency conversion compensation coefficients, through corresponding strategy branching calculations. The instruction package is then injected into the underlying controller via the industrial fieldbus protocol to complete the entire closed-loop process from intelligent diagnosis to physical execution. This allows the entire monitoring method to go beyond the state recognition level and extend to the physical intervention level, forming a complete acquisition-computation-execution technology closed loop. This upgrades the membrane module's operation and maintenance decisions from a passive response mode relying on human experience to an adaptive proactive intervention mode based on real-time state perception.
[0081] More specifically, in the embodiments of this application, step S5 includes: based on the enumerated classification state of the dominant membrane state definition result, logarithmic dose calculation or cleaning lockout flag generation and remaining lifetime estimation are performed using the component reversible contamination index and the component irreversible aging index to obtain chemical cleaning and hardware lockout intervention parameters; based on the preset variable frequency control compensation optimization space, variable frequency drive PID adaptive compensation optimization is performed on the component reversible contamination index and the component irreversible aging index to obtain variable frequency proportional integral derivative compensation parameters; and industrial bus instruction set serialization encapsulation is performed on the chemical cleaning and hardware lockout intervention parameters and the variable frequency proportional integral derivative compensation parameters to obtain an adaptive closed-loop control instruction set.
[0082] Specifically, in a concrete example of this application, the system first enters the corresponding strategy branch based on the enumerated classification state of the dominant membrane state definition result, and calculates targeted intervention parameters using the component reversible fouling index and the component irreversible aging index, respectively. The system parses the enumerated classification code field in the dominant membrane state definition result. If the code is 0x01, indicating a reversible fouling dominant state, the system enters the logarithmic calculation branch for chemical cleaning dosage: the system extracts the current value of the component reversible fouling index, uses its numerical intensity to substitute into the logarithmic growth calculation model to dynamically calculate the optimal dosage of the in-situ chemical cleaning fluid required, and simultaneously sets the hardware cleaning lockout flag to zero, thus allowing the cleaning system to execute. The mathematical expression of this logarithmic dosage calculation is:
[0083] in, The optimal injection dose of the chemical cleaning fluid, calculated in liters, constitutes one of the core control parameters in the chemical cleaning and hardware interlocking intervention parameters. To determine the dosage constant of the single basic cleaning fluid for the current membrane module model, this value is determined by the membrane housing volume and the cleaning fluid concentration specification. In this example, the value is 150 liters for the 8-inch 6-core membrane housing system. This represents the component's reversible contamination index value during the current flaw detection cycle. The value is 1.2, representing the logarithmic sensitivity gain coefficient of the cleaning agent to clogging resistance. In this example, the value is 1.2. In this expression, The design consideration of using logarithmic functions instead of nonlinear functions is that: in the stage of mild fouling, the dosage of cleaning fluid increases approximately linearly with the exponential function to ensure sufficient dissolution kinetic driving force, while in the stage of severe fouling, the saturation compression characteristics of the logarithmic function make the dosage increase tend to be slow, preventing excessive reagent residue in the ultrapure water pipeline from causing irreversible chemical pollution to the subsequent continuous electro-desalination module and terminal polishing resin. The structure multiplied by the correction factor within square brackets ensures that the total cleaning fluid volume is always proportionally adjusted based on the baseline dosage calibrated for the membrane housing volume, rather than being a baseless calculation without physical constraints. If the enumeration code is 0x02, indicating an irreversible aging-dominated state, the system enters the cleaning lockout and lifespan estimation branch: the system sets the hardware cleaning lockout flag to 1, thereby fundamentally blocking the physical execution path of the cleaning system to prevent secondary damage to the degraded membrane surface caused by ineffective chemical cleaning. Simultaneously, it extracts the current value of the irreversible aging index of the module and estimates the remaining number of flaw detection cycles required for the membrane module to progress from the current aging index to full scale 100 (complete failure) based on a linear extrapolation model, and generates a module replacement early warning work order accordingly. If the enumeration code is 0x04, indicating a coupled accelerated deterioration state, the system simultaneously activates the cleaning sequence and the pressure reduction operation command, outputs the cleaning fluid dosage according to the solution result of the reversible fouling-dominated branch, and lowers the transmembrane pressure differential operation setpoint to 85% of the current value to suppress concentration polarization intensity, and generates the highest priority spare parts procurement work order. The system merges and encapsulates the cleaning fluid dosage, lockout flag, lifespan work order, and pressure reduction instruction obtained from the above branches into chemical cleaning and hardware lockout intervention parameters according to the memory structure format.
[0084] Next, the system enters the variable frequency drive PID adaptive compensation optimization stage. Regardless of the enumeration state of the dominant membrane state definition result, the increase in permeability resistance or micro-leakage caused by membrane module performance degradation will cause a shift in the transmembrane fluid inertial time constant of the ultrapure water system. If constant variable frequency drive PID parameters are continued, it is very easy to cause oscillation or overshoot in the pipeline pressure. Based on the preset variable frequency control compensation optimization space, the system performs joint derating calculations on the reversible fouling index and the irreversible aging index of the module to adaptively adjust the proportional gain term of the high-pressure pump frequency converter. The mathematical expression of this PID compensation optimization is:
[0085] in, The new proportional gain coefficient obtained after attenuation compensation calculation constitutes the core parameter in the frequency conversion proportional-integral-derivative compensation parameters. This is the initial value of the reference proportional gain determined during the commissioning of the ultrapure water system when the equipment is in full health condition. In this example, this value is the PID proportional gain value calibrated during the factory commissioning of the high-pressure pump frequency converter. The component's reversible pollution index; The irreversible aging index of the component; The first-order attenuation weighting coefficient is used to measure the sluggishness of the frequency conversion response due to viscoelastic drag. In this example, the value is 0.3. Its physical meaning is that the thickening effect of the reversible fouling layer on the inertia of the transmembrane fluid causes the system response to be sluggish, so the proportional gain needs to be reduced to avoid overshoot. This is the second-order nonlinear attenuation weighting coefficient for the loss of hydraulic damping due to rigid leakage, which is taken as 0.5 in this example. In this expression, The first-order linear derating of the reversible fouling index of the component is applied to the part, which reflects the physical phenomenon that the thickening of the reversible fouling layer leads to an increase in the viscous resistance of the membrane fluid, which in turn slows down the system pressure response. The proportional gain needs to be reduced accordingly to avoid the controller over-correcting on the slow-responding controlled object. A second-order nonlinear derating is applied to the irreversible aging index of some components. The physical consideration for choosing a quadratic term instead of a linear term is that the micropore size expansion caused by irreversible aging has a sublinear impact on the hydraulic characteristics of the system in the early stages, but its destructive effect on fluid damping accelerates as aging progresses. The quadratic term structure can more accurately fit this nonlinear deterioration curve, ensuring that the proportional gain is only slightly adjusted during mild aging and significantly drated during severe aging to maintain system stability. After the proportional gain calculation is completed, the integral time constant and derivative time constant are adaptively adjusted according to the same design concept, and the three parameters are combined and encapsulated into a variable frequency proportional-integral-derivative compensation parameter.
[0086] Then, the chemical cleaning and hardware interlocking intervention parameters and the frequency conversion proportional-integral-derivative compensation parameters are serialized and encapsulated using the industrial bus instruction set. Based on the fieldbus communication protocol of the underlying Siemens programmable logic controller of the ultrapure water treatment device, the system performs word length padding, checksum calculation, and frame header delimitation serialization processing on the above two types of heterogeneous control parameters. Specifically, the system encodes the chemical cleaning fluid dosage value into the cleaning dosage register address in the Modbus register mapping table according to a 4-byte single-precision floating-point format, writes the interlocking flag into the discrete output coil address according to a 1-bit Boolean value, and encodes the frequency conversion proportional gain compensation value into the frequency converter PID parameter register address according to the data format agreed upon by the frequency converter communication protocol. All parameters are then serialized according to the predefined register address order and supplemented with a cyclic redundancy check code to form a complete communication frame message. The system then packages the serialized messages into an adaptive closed-loop control instruction set.
[0087] Furthermore, the adaptive closed-loop control instruction set is sent to the programmable logic controller (PLC) for execution via the Modbus TCP industrial Ethernet interface. Upon receiving the instruction set, the PLC parses the control parameters corresponding to each register address and maps them to the physical output ports in its next scan cycle: the value of the cleaning dosage register is converted into the operating time and frequency setpoint of the dosing pump to control the precise injection volume of the cleaning fluid; the interlock flag is mapped to the relay output channel of the main valve of the cleaning system to control its physical on / off state; and the variable frequency PID compensation parameters are written into the parameter register of the high-pressure pump inverter, overwriting its proportional gain setpoint during operation. After the instruction is sent, the system initiates a response timeout monitoring mechanism, waiting for the execution confirmation message returned by the PLC. If no confirmation is received within the preset timeout window, a communication fault alarm is triggered, and the instruction set is cached in the retransmission queue. The entire global closed loop from intelligent diagnostic assessment to physical execution intervention is thus completed. Following the preset flaw detection execution cycle, the system repeats all the above steps after the next perturbation flaw detection every 4 hours, continuously outputting an adaptive control strategy that matches the real-time health status of the membrane module.
[0088] Furthermore, an intelligent monitoring system for the operating status of an ultrapure water treatment device is also provided.
[0089] Figure 6 This is a block diagram of an intelligent monitoring system for the operating status of an ultrapure water treatment apparatus according to an embodiment of this application. Figure 6 As shown, the intelligent monitoring system 100 for the operation status of an ultrapure water treatment device according to an embodiment of this application includes: a hydraulic perturbation excitation injection and data acquisition module 110, used to inject a hydraulic step perturbation signal of preset frequency amplitude into the reverse osmosis high-pressure pump through a frequency converter drive when the ultrapure water treatment device is in steady-state water production operation, and synchronously acquire multi-source transient sensor data at high frequency and perform clock alignment and splicing to obtain transient hydraulic response time series data; a perturbation response dynamics feature extraction module 120, used to extract perturbation response dynamics features from the transient hydraulic response time series data to obtain the viscoelastic response feature matrix of the membrane module; and a membrane physical property decoupling and exponential mapping module 130, used to decouple the membrane module physical properties and perform exponential mapping. The viscoelastic response feature matrix of the component is used to decouple the physical properties of the membrane and perform exponential mapping to obtain the component reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the component irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. The dual-track attenuation state comprehensive evaluation module 140 is used to perform a dual-track attenuation state comprehensive evaluation on the component reversible fouling index and the component irreversible aging index to obtain the dominant membrane state definition result. The strategy optimization and control instruction generation module 150 is used to perform branch strategy optimization on the dominant membrane state definition result, the component reversible fouling index, and the component irreversible aging index to generate an adaptive closed-loop control instruction set, which is then sent to the programmable logic controller for execution.
[0090] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.
Claims
1. A method for intelligent monitoring of the operating status of an ultrapure water treatment device, characterized in that, include: S1, when the ultrapure water treatment device is in steady-state water production operation, a hydraulic step perturbation signal with a preset frequency amplitude is injected into the reverse osmosis high-pressure pump through the frequency conversion drive end, and multi-source transient sensor data is simultaneously collected at high frequency and clock-aligned and spliced to obtain transient hydraulic response time series data. S2, perform perturbation response dynamics feature extraction on transient hydraulic response time series data to obtain the viscoelastic response feature matrix of membrane module; S3, the membrane physical properties of the membrane module are decoupled and indexed by the membrane module viscoelastic response characteristic matrix to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. S4, a dual-track degradation state comprehensive evaluation of the component's reversible pollution index and irreversible aging index is conducted to obtain the dominant membrane state definition result. S5 optimizes the branching strategy based on the dominant membrane state definition results, the component reversible contamination index, and the component irreversible aging index to generate an adaptive closed-loop control instruction set, which is then sent to the programmable logic controller for execution.
2. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 1, characterized in that, Multi-source transient sensor data includes reverse osmosis feed water pressure signal, product water flow signal, and product water conductivity signal.
3. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 1, characterized in that, Step S1 includes: After confirming that the ultrapure water treatment unit is in steady-state water production operation, a step signal model is performed by superimposing a preset frequency offset on the high-pressure pump reference operating frequency data to obtain the step perturbation control signal. The step perturbation control signal is injected into the drive end of the reverse osmosis high-pressure pump by a frequency converter to excite transient hydraulic disturbance. At the injection time, the observation time window is opened simultaneously to extract the pressure signal at the reverse osmosis feed water end, the flow signal at the product water end, and the conductivity signal at the product water end by a fixed-length sequence to obtain a multi-source transient sensing dataset. Equal-interval global clock resampling alignment and multi-channel column vector concatenation are performed on the asynchronous sampling channels in the multi-source transient sensing dataset caused by differences in hardware communication cycles to obtain transient hydraulic response time series data.
4. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 1, characterized in that, Step S2 includes: Cross-correlation phase hysteresis characteristic calculation and data demultiplexing are performed on transient hydraulic response time series data to obtain phase hysteresis characteristic sequence and residual time series data of pressure conductivity; Synchronous instantaneous rate of change derivatives are performed on the high-frequency pressure sequence and high-frequency conductivity sequence in the residual time series data of pressure conductivity, and the ratio sequence of conductivity rate of change to pressure rate of change is calculated to obtain the transient response gradient characteristic sequence. The phase hysteresis feature sequence and the transient response gradient feature sequence are orthogonally concatenated in the column dimension to obtain the viscoelastic response feature matrix of the membrane module.
5. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 1, characterized in that, Step S3 includes: Based on the feature channel indexing mechanism, the viscoelastic response feature matrix of the membrane module is orthogonally sliced and decoupled in the column dimension to obtain the first channel data characterizing the dynamic delay characteristics and the second channel data characterizing the mass transfer agility characteristics, thus obtaining the smooth phase hysteresis feature set and the smooth response gradient feature set. The component reversible contamination index is obtained by performing porous medium viscoelastic deformation mapping and reversible contamination quantification on the smooth phase hysteresis feature set. The component's irreversible aging index is obtained by mapping the smooth response gradient feature set to a microporous rigid expansion model and quantifying irreversible aging.
6. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 1, characterized in that, Step S4 includes: The reversible contamination index and irreversible aging index of the component are compared with the preset reversible contamination cleaning critical threshold and irreversible chemical failure critical threshold in the system knowledge base to obtain the reversible physical resistance deviation and irreversible chemical damage deviation. Two-dimensional attenuation contribution ratios were quantified for the deviation of reversible physical resistance and the deviation of irreversible chemical damage to obtain the contribution ratios of pollution attenuation and aging attenuation. Based on the operation and maintenance strategy classification decision tree model, the contribution ratios of pollution attenuation and aging attenuation are compared with the preset absolute dominant boundary and enumeration classification inference is completed to obtain the dominant membrane state definition result.
7. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 1, characterized in that, Step S5 includes: Based on the enumerated classification state of the dominant membrane state definition results, logarithmic dose calculation or cleaning lockout flag generation and remaining lifetime estimation are performed using the component reversible pollution index and the component irreversible aging index to obtain chemical cleaning and hardware lockout intervention parameters. Based on the preset variable frequency control compensation optimization space, variable frequency drive PID adaptive compensation optimization is performed on the component reversible pollution index and the component irreversible aging index to obtain the variable frequency proportional integral derivative compensation parameters. The parameters for chemical cleaning and hardware interlocking intervention, as well as the parameters for frequency conversion proportional-integral-derivative compensation, are serialized and encapsulated using an industrial bus instruction set to obtain an adaptive closed-loop control instruction set.
8. The intelligent monitoring method for the operating status of the ultrapure water treatment device according to claim 6, characterized in that, Based on the operation and maintenance strategy classification decision tree model, the contribution ratios of pollution attenuation and aging attenuation are compared with the preset absolute dominant boundary, and enumeration classification inference is performed to obtain the dominant membrane state definition results, including: The coupling synergy strength of the two degradation channels, reversible physical resistance deviation and irreversible chemical damage deviation, is calculated to obtain the pollution aging coupling synergy factor. Based on the pollution-aging coupling synergy factor, the contribution ratios of pollution attenuation and aging attenuation are corrected by coupling gain and the basic dominant boundary is compressed to obtain the coupled corrected pollution contribution ratio, the coupled corrected aging contribution ratio, and the adaptive dominant judgment boundary. Multi-path condition comparison and enumeration classification inference were performed on the proportion of pollution contribution, the proportion of aging contribution, the adaptive dominant determination boundary, and the pollution-aging coupling synergy factor to obtain the dominant membrane state definition results.
9. An intelligent monitoring system for the operating status of an ultrapure water treatment device, characterized in that, include: The hydraulic perturbation excitation injection and data acquisition module is used to inject a hydraulic step perturbation signal of preset frequency amplitude into the reverse osmosis high-pressure pump through the frequency conversion drive end when the ultrapure water treatment device is in steady-state water production operation. Simultaneously, it acquires multi-source transient sensor data at high frequency and performs clock alignment and splicing to obtain transient hydraulic response time series data. The perturbation response dynamics feature extraction module is used to extract perturbation response dynamics features from transient hydraulic response time-series data to obtain the viscoelastic response feature matrix of the membrane module. The membrane physical property decoupling and exponential mapping module is used to decouple and map the membrane physical properties of the membrane module viscoelastic response characteristic matrix to obtain the module reversible fouling index, which characterizes the degree of reversible fouling accumulation, and the module irreversible aging index, which characterizes the degree of chemical degradation of the membrane substrate. The dual-track degradation state comprehensive evaluation module is used to comprehensively evaluate the reversible pollution index and the irreversible aging index of the module to obtain the dominant membrane state definition result. The strategy optimization and control instruction generation module is used to perform branch strategy optimization on the dominant membrane state definition result, the component reversible pollution index and the component irreversible aging index to generate an adaptive closed-loop control instruction set, which is then sent to the programmable logic controller for execution.
10. The intelligent monitoring system for the operating status of the ultrapure water treatment device according to claim 9, characterized in that, The dual-track attenuation state comprehensive evaluation module includes: The critical threshold deviation calculation unit is used to compare the component reversible pollution index and the component irreversible aging index with the preset reversible pollution cleaning critical threshold and irreversible chemical failure critical threshold in the system knowledge base to obtain the reversible physical resistance deviation and the irreversible chemical damage deviation. The two-dimensional attenuation contribution ratio quantification unit is used to quantify the deviation of reversible physical resistance and the deviation of irreversible chemical damage in two dimensions to obtain the pollution attenuation contribution ratio and the aging attenuation contribution ratio. The dominant membrane state decision inference unit is used to compare the contribution ratios of pollution attenuation and aging attenuation with the preset absolute dominant boundary based on the operation and maintenance strategy classification decision tree model, and complete the enumeration classification inference to obtain the dominant membrane state definition result.
