A developing solution recovery system abnormality early warning method

Abstract

This invention discloses an abnormal early warning method for a developer recovery system. The method involves real-time acquisition of developer concentration signals, followed by amplification, low-pass filtering, and analog-to-digital conversion preprocessing to obtain digital concentration values. A sliding window least squares method is used to calculate the concentration slope value. Simultaneously, using the average concentration value after a preset period of stable system operation as a benchmark, a first-order low-pass digital filter is used to extract the steady-state concentration component, and the steady-state concentration drift is calculated. The concentration slope value is compared with a preset slope mutation threshold, and the steady-state concentration drift is compared with a preset steady-state drift amplitude threshold. Based on the comparison results, graded early warnings are output for sensor failure risk, pipeline blockage risk, and combined severe faults.

Description

Technical Field

[0001] This invention relates to an abnormality early warning method for a developer recovery system, belonging to the technical field of developer recovery equipment. Background Technology

[0002] In semiconductor and flat panel display manufacturing processes, the concentration control of the developer directly affects the accuracy of photolithography patterns and product yield. The developer recovery system achieves the recycling of the developer through online monitoring and regeneration. Among these components, the concentration sensor, which monitors the concentration of the recycled developer in real time, is a key component ensuring accurate concentration control.

[0003] The existing developer recovery system mainly uses the following methods for abnormal monitoring: a concentration sensor is installed in the recovery pipeline, and a fixed threshold over-limit alarm is used as the core fault identification strategy. That is, when the measured concentration value exceeds the preset upper or lower limit, an alarm is triggered. At the same time, regular manual inspections are carried out, with operators checking the sensor zero point and observing the pipeline condition on a shift basis. The system immediately shuts down after triggering the threshold alarm, or the operator manually performs an emergency shutdown.

[0004] However, the aforementioned prior art has the following technical defects:

[0005] First, zero-point drift and response decay of sensors are difficult to detect in a timely manner. After long-term operation, concentration sensors may experience zero-point drift or sensitivity decay, but their output values ​​may still remain within a fixed threshold range. This can lead to the actual concentration deviating from the process requirements without the system indicating it, ultimately resulting in poor batch development.

[0006] Second, there are few signs in the early stages of pipeline crystallization blockage. The gradual deposition of crystals or particles in the recovery pipeline will cause a slow concentration drift, but because it does not trigger an absolute concentration exceeding the limit, it cannot be detected by the system until the blockage becomes completely severe, forcing the system to shut down urgently.

[0007] Third, the fixed threshold alarm is severely delayed. An alarm is only generated when the concentration exceeds the fixed upper and lower limits, failing to capture abnormal trends such as sudden changes in the slope of the concentration process curve or steady-state drift. By this time, the fault has already occurred, and the system has lost the opportunity for early intervention.

[0008] Fourth, manual inspections lack real-time capability. Sudden anomalies may occur during the time interval between two inspections, causing equipment to operate with defects for extended periods. Furthermore, manual troubleshooting cannot quickly pinpoint the root cause of the fault, resulting in low maintenance efficiency.

[0009] Fifth, unplanned downtime results in significant losses. The existing alarm-based shutdown strategy lacks the buffer capacity to reprocess the current batch. Once an alarm is triggered, production is directly interrupted, leading to the scrapping of the current batch of products and increased production line reset costs.

[0010] Sixth, false alarms and missed alarms coexist. Fixed thresholds cannot adapt to the dynamic characteristics of concentration curves under different volumes and different operating stages, which easily leads to false alarms or missed alarms, reducing operators' trust in the alarm system.

[0011] Therefore, there is an urgent need for an anomaly early warning method that can perform predictive maintenance based on the dynamic characteristics of the concentration process curve, so as to identify abnormal trends in advance before sensor failure or pipeline blockage occurs, thereby reducing the risk of sudden shutdown and ensuring production continuity. Summary of the Invention

[0012] The present invention provides an abnormal early warning method for a developer recovery system in order to solve the problems existing in the prior art.

[0013] The technical solutions adopted in this invention are as follows:

[0014] A method for early warning of abnormalities in a developer recovery system includes the following steps:

[0015] S1: Real-time acquisition of developer concentration signal to obtain digital concentration value;

[0016] S2: Calculate the concentration slope value based on the time series data of the digital concentration value, extract the steady-state component of the concentration based on the time series data of the digital concentration value, and calculate the absolute value of the difference between the steady-state component of the concentration and the reference value as the steady-state drift of the concentration.

[0017] S3: Compare the concentration slope value with the preset slope abrupt change threshold, and compare the concentration steady-state drift with the preset steady-state drift amplitude threshold;

[0018] S4: Output the corresponding warning signal based on the comparison results, where:

[0019] When the absolute value of the concentration slope exceeds the slope mutation threshold and the steady-state drift of the concentration does not exceed the steady-state drift amplitude threshold, a sensor failure risk warning signal is output.

[0020] When the steady-state drift of concentration exceeds the steady-state drift amplitude threshold and the absolute value of the concentration slope does not exceed the slope change threshold, a pipeline blockage risk warning signal is output.

[0021] When the absolute value of the concentration slope exceeds the slope mutation threshold and the steady-state drift of the concentration exceeds the steady-state drift amplitude threshold, a composite severe fault warning signal is output.

[0022] Furthermore, in S1, the developer concentration signal is sequentially amplified, low-pass filtered, and converted from analog to digital to obtain a digital concentration value.

[0023] Further, in S2, calculating the concentration slope value includes: calculating the concentration slope value using linear fitting with a sliding window least squares method;

[0024] The calculation of steady-state concentration drift includes: using the average value of digital concentration values ​​collected during a preset period of startup and stable operation of the developer recovery system as the reference value, using a first-order low-pass digital filter to obtain the current steady-state concentration component, and calculating the absolute value of the difference between the current steady-state concentration component and the reference value as the steady-state concentration drift.

[0025] Furthermore, the slope mutation threshold is ±0.05% / second, and the steady-state drift amplitude threshold is 0.20%.

[0026] Furthermore, in S4, the output of the composite critical fault warning signal includes driving the audible and visual alarm to emit an audible and visual alarm;

[0027] The output sensor failure risk warning signal includes triggering the sensor failure indication circuit;

[0028] The output pipeline blockage risk warning signal includes triggering the pipeline blockage detection circuit.

[0029] Furthermore, it also includes step S5: calculating the comprehensive health score based on multi-dimensional degradation indicators, wherein the comprehensive health score is calculated using the following formula:

[0030] ,

[0031] in, Let i be the weight of the i-th degenerate dimension. Let be the normalized degradation value of the i-th degradation dimension. Initial value for overall health .

[0032] Furthermore, the normalized degenerate value Calculate using the following formula:

[0033] ,

[0034] in, For the original observation of the i-th degenerate dimension, This represents the historical maximum value of the original observations. This is the historical minimum value of the original observation.

[0035] Furthermore, it also includes step S6: based on the historical sequence of comprehensive health, the remaining effective life is predicted using the exponential decay model and the Weibull reliability model respectively, and the model with higher fitting degree is selected to output the predicted value of the remaining effective life.

[0036] The exponential decay model is as follows:

[0037] ,

[0038] The Weibull reliability model is as follows:

[0039] ,

[0040] in, The exponential decay coefficient is... For the Weibull scale parameter, These are Weibull shape parameters.

[0041] Furthermore, it also includes step S7:

[0042] When any level of warning signal is triggered, a delayed shutdown request is sent. The delay time of the delayed shutdown request is the sum of the remaining time of the current batch and the fixed buffer time. If the warning signal disappears within the delay time, the delayed shutdown request is cancelled. If the warning signal continues until the end of the delay time, shutdown is performed.

[0043] Furthermore, it also includes step S8:

[0044] The warning events are stored in a non-volatile storage area. The warning events include the trigger time, the anomaly type, the concentration slope value at the time of triggering, and the concentration steady-state drift.

[0045] When a warning reset command is received, the current warning status is cleared. If the root cause of the fault is not eliminated, the warning signal is triggered again within a preset time.

[0046] The present invention has the following beneficial effects:

[0047] (1) By calculating the slope value of the concentration process curve in real time and monitoring the slope change, an early warning can be issued in the early stage of sensor signal jump, electrode contamination or poor contact, so as to avoid the concentration control failure caused by the slow failure of the sensor.

[0048] (2) Extract the steady-state concentration component by low-pass filtering and compare it with the benchmark value. When the concentration slowly drifts due to pipeline crystallization blockage, increased filter pressure loss, etc., alarm in time to prevent the blockage from completely deteriorating and causing an emergency shutdown.

[0049] (3) By comparing the concentration slope value and the concentration steady-state drift with the corresponding threshold, a mapping relationship is established between abnormal slope and sensor failure, and abnormal drift and pipeline blockage. Maintenance personnel can directly locate the fault type without comprehensive investigation.

[0050] (4) Significantly improves the real-time performance of anomaly monitoring, eliminates blind spots in inspection, and automatically and continuously collects concentration data and determines anomalies in real time. It can monitor equipment status around the clock without relying on manual inspection.

[0051] (5) Effectively reduce production losses caused by unplanned shutdowns. A graded early warning mechanism is used to issue a warning in the early stage of a fault, allowing operators time to intervene. In the preferred embodiment, a delayed shutdown strategy can be further configured to allow shutdown to be performed after the current batch of production is completed, thus avoiding batch scrapping caused by direct interruption of production.

[0052] (6) By replacing the single fixed threshold judgment with dynamic feature monitoring, it can adapt to the concentration curve characteristics of different volumes and different operating stages, and filter instantaneous interference by combining the abnormal duration judgment, thereby reducing the probability of false alarms and missed alarms. Attached Figure Description

[0053] Figure 1 This is a block diagram illustrating the principle of the present invention.

[0054] Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0055] The invention will now be further described with reference to the accompanying drawings.

[0056] This embodiment is applied to a developer recovery system in semiconductor and flat panel display manufacturing processes. It is used to provide early warning of abnormal operating conditions such as sensor performance degradation and pipeline crystallization blockage in the developer recovery pipeline, thereby enabling predictive maintenance and reducing the risk of unplanned downtime.

[0057] like Figure 1 The developer recovery system's recovery pipeline is equipped with a concentration sensor. The signal output of the concentration sensor is connected to the input of a signal acquisition module, and the digital signal output of the signal acquisition module is connected to a microcontroller unit. The microcontroller unit is bidirectionally connected to an anomaly feature database storage unit and an early warning trigger module. The output of the early warning trigger module is connected to a pipeline blockage detection circuit, a sensor failure indication circuit, and an audible and visual alarm. The microcontroller unit also interacts with the main controller of the developer recovery system via a communication interface. The main controller is connected to a shutdown actuator to receive early warning signals and control the shutdown logic.

[0058] Combination Figure 2 The anomaly warning method executed by the above system specifically includes the following process.

[0059] The concentration sensor collects the concentration of the developer flowing through the recovery pipeline in real time and outputs a raw analog voltage signal. In this embodiment, the concentration sensor is a conductivity type concentration sensor with an output signal range of 0-10mV, corresponding to a developer concentration range of 0-5%.

[0060] After receiving the raw analog signal, the signal acquisition module sequentially performs amplification, low-pass filtering, and analog-to-digital conversion. The amplification gain is set to 10 times, expanding the raw signal from 0-10mV to the 0-100mV range to match the input range of the subsequent analog-to-digital conversion. The low-pass filtering uses a first-order RC low-pass filter circuit with a cutoff frequency of 0.5Hz to filter out power frequency interference and high-frequency noise caused by fluid fluctuations in the pipeline, while retaining the effective trend signal of concentration changes. The analog-to-digital conversion uses a 12-bit resolution converter, with a conversion range corresponding to the concentration range of 0-5%, outputting the corresponding digital concentration value after conversion.

[0061] The microcontroller reads the converted digital concentration value at a sampling frequency of 1Hz and stores it in the internal data buffer. The buffer has a storage capacity of no less than 300 data points and can continuously store 5 minutes of time-series digital concentration values, providing a data foundation for subsequent feature calculations.

[0062] The microcontroller unit calculates the concentration slope of the concentration curve in real time based on cached time-series digital concentration values. In this embodiment, a sliding window least squares linear fitting method is used to calculate the slope, with a calculation interval of 10 seconds, meaning the current concentration slope value is updated every 10 seconds. For each calculation, all digital concentration values ​​within the 30 seconds prior to the current moment are used as the calculation window, containing 30 sampling data points. The relative sampling time is used as the independent variable, and the corresponding digital concentration value is used as the dependent variable. A univariate linear fitting is performed using the least squares method to obtain the slope of the fitted line, which is the concentration slope value at the current moment, expressed in % / second. The formula for least squares fitting is:

[0063] ,

[0064] Where n is the total number of sampling points in the sliding window, and in this embodiment n=30; This represents the relative time value corresponding to the i-th sampling point, in seconds; This represents the digital concentration value corresponding to the i-th sampling point, in units of % This represents the concentration slope value at the current moment. The calculated concentration slope value is stored in a circular queue of length 60, which can continuously store 10 minutes of historical slope data for subsequent continuous anomaly detection.

[0065] The microcontroller unit synchronously calculates the steady-state concentration drift based on time-series digital concentration values ​​to identify long-term, slow concentration shift trends. The calculation process consists of three stages: baseline value determination, steady-state component extraction, and drift calculation. First, an initial baseline value is determined. After the developer recovery system starts up, completes self-test, and operates stably for 30 minutes, the microcontroller unit calculates the arithmetic mean of all digital concentration values ​​within those 30 minutes and sets this average value as the initial baseline value. This serves as a reference for subsequent drift determination. A first-order low-pass digital filter algorithm is then used to smooth the real-time digital concentration value, extracting the steady-state component of the concentration signal and filtering out transient fluctuations. The recursive formula for the first-order low-pass digital filter is:

[0066] ,

[0067] in, This is the digital concentration value at the current moment, in % . The steady-state concentration component at the previous moment is expressed in % (%). For the filter coefficients, in this embodiment The value is set to 0.01, corresponding to a time constant of approximately 100 seconds, which can effectively smooth short-term fluctuations and track long-term concentration trends. The initial value for the filter recursion is assigned using the first concentration value after the system stabilizes.

[0068] After obtaining the steady-state concentration component at the current moment, calculate the absolute value of the difference between it and the initial reference value; this is the steady-state concentration drift at the current moment. The calculation formula is:

[0069] ,

[0070] The unit for steady-state concentration drift is %, which is used to characterize the long-term deviation of the current concentration from the reference value.

[0071] The microcontroller reads preset threshold parameters from the anomaly feature library storage unit and compares the real-time calculated concentration slope value and steady-state concentration drift with the corresponding thresholds to determine whether the current system has an anomaly and the type of anomaly. The threshold parameters pre-stored in the anomaly feature library storage unit include a slope mutation threshold and a steady-state drift amplitude threshold. In this embodiment, the slope mutation threshold is set to ±0.05% / second, and the steady-state drift amplitude threshold is set to 0.20%. These threshold parameters can be adjusted online according to the developer type, total system volume, and production process requirements to adapt to different operating conditions. The specific determination logic is as follows:

[0072] If the absolute value of the concentration slope is continuously greater than the slope change threshold for 30 consecutive seconds, and the steady-state drift of the concentration does not exceed the steady-state drift amplitude threshold, then it is determined that there is a risk of sensor failure. The cause of failure is usually sensor signal jump, electrode contamination, or poor contact of the transmission line.

[0073] If the steady-state concentration drift is consistently greater than the steady-state drift amplitude threshold for 30 consecutive seconds, and the absolute value of the concentration slope does not exceed the slope mutation threshold, then it is determined that there is a risk of pipeline blockage. The cause of the failure is usually pipeline crystal deposition, increased filter pressure loss, or decreased pump efficiency.

[0074] If both of the above-mentioned abnormalities occur simultaneously and last for more than 60 seconds, it is determined that there is a combined serious fault, corresponding to the condition of complete sensor failure or severe pipeline blockage.

[0075] Based on the fault determination result, the microcontroller sends the corresponding control command to the early warning trigger module, which then outputs early warning signals in stages.

[0076] When a risk of pipeline blockage is detected, the early warning trigger module outputs the first control signal, triggering the pipeline blockage detection circuit and illuminating the yellow indicator light to alert maintenance personnel to the trend of pipeline blockage.

[0077] When a sensor failure risk is detected, the early warning trigger module outputs a second control signal to trigger the sensor failure indication circuit, illuminating an orange indicator light to alert maintenance personnel to the abnormal sensor performance.

[0078] When a complex and serious fault is detected, the early warning trigger module simultaneously outputs three control signals. While illuminating the yellow and orange indicator lights, it drives the audible and visual alarm 8 to start, emitting an 85dB buzzer sound and flashing red light to provide the highest level of fault warning.

[0079] The microcontroller unit is also connected to a warning signal suppression shutdown module, which is connected to the shutdown enable signal of the main controller. When any level of warning signal is triggered, the warning signal suppression shutdown module captures the warning signal and sends a delayed shutdown request to the main controller. The duration of the delayed shutdown is calculated in real time by the main controller, specifically the sum of the remaining processing time of the current production batch and a fixed buffer time of 30 seconds. The remaining processing time of the current batch is calculated by the main controller based on the volume of developed solution already processed and the real-time flow data of the pipeline. If the warning signal disappears automatically within the delay period, the previous anomaly is determined to be a false trigger caused by transient interference. The warning signal suppression shutdown module sends a command to the main controller to cancel the delayed shutdown, and the system maintains normal operation without performing a shutdown operation. If the warning signal persists until the end of the delay period, the main controller releases the shutdown restriction, sends a shutdown command to the shutdown execution mechanism, and performs an orderly shutdown after completing the current batch of production to avoid interrupting production and causing batch product scrapping.

[0080] All triggered warning events are automatically recorded and stored in the non-volatile storage area of ​​the anomaly feature database storage unit. Each warning record includes the warning trigger time, anomaly type, concentration slope value at the trigger time, and concentration steady-state drift data. The anomaly feature database storage unit can store up to 1000 historical warning records. When the storage capacity is full, the oldest historical record is automatically overwritten, achieving cyclic storage. The control cabinet panel of the developer recovery system is equipped with a warning reset button. After maintenance personnel have investigated and eliminated the root cause of the fault, pressing the warning reset button will trigger the microcontroller to receive the warning reset command, clear the currently active warning status, and restore the indicator lights and audible and visual alarms to normal. If the root cause of the fault is not completely eliminated, the system will detect the anomaly again within 10 seconds after the reset and re-trigger the corresponding level of warning signal.

[0081] The microcontroller unit also features sensor health assessment and remaining effective lifespan prediction capabilities. Through the fusion calculation of multi-dimensional degradation indicators, it achieves an upgrade from qualitative early warning to quantitative maintenance guidance. The specific implementation process is as follows:

[0082] First, raw observations of four degradation indicators were collected: zero-point drift rate, response delay increment, signal-to-noise ratio degradation, and linearity deviation. The zero-point drift rate was calculated based on the change in the baseline value over time, representing the shift in baseline concentration per unit time, expressed as % / day. The response delay increment was calculated by extracting the concentration step rise time during system mode switching, including natural step scenarios such as switching between developer recovery and regeneration modes and replenishment of new solution. The difference between the current step rise time and the baseline rise time set during the initial factory calibration was used to obtain the response delay increment, expressed in seconds. The signal-to-noise ratio degradation was obtained by estimating the power spectral density of the time-series digital concentration values. The Welch method was used to calculate the power spectrum of the concentration signal within the effective bandwidth of 0~0.5Hz. The signal power in the low-frequency band of 0~0.1Hz and the high-frequency band of 0.2~0.5Hz were statistically analyzed, and the power ratio of the two frequency bands was calculated to characterize the degree of signal-to-noise ratio degradation. The linearity deviation was obtained from periodic offline calibration data. Maintenance personnel calibrated at multiple points by introducing standard concentration solutions according to the maintenance cycle, and recorded the maximum deviation between the sensor measured value and the standard value at each concentration point. The calibration results were stored in the abnormal feature database storage unit, and the latest calibration data was called when calculating the health status.

[0083] After obtaining the original observations for each dimension, the observations for each dimension are normalized, mapping the indicators of different dimensions to the 0-1 interval, thus obtaining the normalized degenerate value for each dimension. The normalization calculation formula is as follows:

[0084] ,

[0085] Where i is the index of the degradation dimension, with a value from 1 to 4; This represents the original observation value of the i-th degenerate dimension at the current moment; The historical maximum value of the original observation value of the i-th degradation dimension corresponds to the most severe degradation state; The historical minimum value of the original observation in the i-th degenerate dimension corresponds to the new state of the sensor; Let be the normalized degradation value of the i-th degradation dimension at the current moment. The value ranges from 0 to 1. The larger the value, the more severe the degradation of the dimension.

[0086] After normalization, the normalized degradation values ​​of the four dimensions are weighted and fused to calculate the overall health of the sensor. The formula for calculating the overall health is as follows:

[0087] ,

[0088] in, is the weight coefficient for the i-th degradation dimension, and the sum of all weight coefficients is 1. It can be configured according to the degree of influence of each dimension on sensor performance. The overall health status at the current moment, with a value ranging from 0 to 1; the initial health status of the sensor in its new state. A lower health score indicates more severe overall sensor degradation. When the overall health score falls below 0.8, the system enters degradation tracking mode, continuously updating the historical sequence of the overall health score to provide a data basis for lifespan prediction.

[0089] When predicting the remaining effective lifetime, both the exponential decay model and the Weibull reliability model are used to fit the historical sequence of overall health. The optimal model is selected for lifetime prediction by comparing the fitting accuracy. The exponential decay model is suitable for scenarios where the sensor degrades slowly and uniformly, and its expression is:

[0090] ,

[0091] in, The exponential decay coefficient represents the rate of health degradation, obtained by taking the logarithm of historical health data and then performing linear regression. The Weibull reliability model is applicable to failure scenarios with progressively accelerating degradation rates, and its expression is:

[0092] ,

[0093] in, is the Weibull scale parameter, representing the feature lifetime; The Weibull shape parameter represents the degradation acceleration characteristic. The degradation rate increases over time; the two parameters are obtained by regression analysis after linearizing the model.

[0094] The least squares residual comparison method was used for model selection. The sum of squared errors between the fitted values ​​of the two models and the actual health values ​​was calculated. A smaller sum of squared errors indicates a higher good fit, and the model with the higher good fit was selected as the final prediction model. After determining the prediction model, the health failure threshold was substituted into the model for inverse kinematics to calculate the time required for the health to drop to the failure threshold, which is the remaining effective lifespan (RUL). In this embodiment, the health failure threshold is set to 0.3, meaning that when the overall health drops to 0.3, the sensor is considered to have reached a state requiring replacement or calibration. Taking the exponential decay model as an example, the formula for calculating the remaining effective lifespan is:

[0095] ,

[0096] in, This is the initial health status of the sensor, which is set to 1 in a brand new state. The system sets the health failure threshold and ultimately outputs the predicted value and confidence interval of the remaining effective lifespan, providing quantitative basis for maintenance personnel to formulate maintenance plans. For example, it may indicate that "the current sensor health is 0.72, and it is expected to drop to the failure threshold in 6.5 days. It is recommended to arrange calibration or replacement within 5 to 7 days."

[0097] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for early warning of abnormalities in a developer recovery system, characterized in that: Includes the following steps: S1: Real-time acquisition of developer concentration signal to obtain digital concentration value; S2: Calculate the concentration slope value based on the time series data of the digital concentration value, extract the steady-state component of the concentration based on the time series data of the digital concentration value, and calculate the absolute value of the difference between the steady-state component of the concentration and the reference value as the steady-state drift of the concentration. S3: Compare the concentration slope value with the preset slope abrupt change threshold, and compare the concentration steady-state drift with the preset steady-state drift amplitude threshold; S4: Output the corresponding warning signal based on the comparison results, where: When the absolute value of the concentration slope exceeds the slope mutation threshold and the steady-state drift of the concentration does not exceed the steady-state drift amplitude threshold, a sensor failure risk warning signal is output. When the steady-state drift of concentration exceeds the steady-state drift amplitude threshold and the absolute value of the concentration slope does not exceed the slope change threshold, a pipeline blockage risk warning signal is output. When the absolute value of the concentration slope exceeds the slope mutation threshold and the steady-state drift of the concentration exceeds the steady-state drift amplitude threshold, a composite severe fault warning signal is output.

2. The abnormal early warning method for the developer recovery system as described in claim 1, characterized in that: In S1, the developer concentration signal is sequentially amplified, low-pass filtered, and converted from analog to digital to obtain a digital concentration value.

3. The abnormal early warning method for the developer recovery system as described in claim 1, characterized in that: In S2, the calculation of the concentration slope value includes: calculating the concentration slope value using linear fitting with the sliding window least squares method; The calculation of steady-state concentration drift includes: using the average value of digital concentration values ​​collected during a preset period of startup and stable operation of the developer recovery system as the reference value, using a first-order low-pass digital filter to obtain the current steady-state concentration component, and calculating the absolute value of the difference between the current steady-state concentration component and the reference value as the steady-state concentration drift.

4. The abnormal early warning method for the developer recovery system as described in claim 1, characterized in that: The slope abrupt change threshold is ±0.05% / second, and the steady-state drift amplitude threshold is 0.20%.

5. The abnormal early warning method for the developer recovery system as described in claim 1, characterized in that: In S4, the output of the composite critical fault early warning signal includes driving the audible and visual alarm to emit an audible and visual alarm. The output sensor failure risk warning signal includes triggering the sensor failure indication circuit; The output pipeline blockage risk warning signal includes triggering the pipeline blockage detection circuit.

6. The abnormal early warning method for the developer recovery system as described in claim 1, characterized in that: The process also includes step S5: calculating the overall health score based on multi-dimensional degradation indicators, wherein the overall health score is calculated using the following formula: ， in, Let i be the weight of the i-th degenerate dimension. Let be the normalized degradation value of the i-th degradation dimension. Initial value for overall health .

7. The abnormal early warning method for the developer recovery system as described in claim 6, characterized in that: The normalized degradation value Calculate using the following formula: ， in, For the original observation of the i-th degenerate dimension, This represents the historical maximum value of the original observations. This is the historical minimum value of the original observation.

8. The developer recovery system anomaly early warning method as described in claim 6, characterized in that: It also includes step S6: Based on the historical sequence of comprehensive health, the remaining effective life is predicted using the exponential decay model and the Weibull reliability model respectively, and the model with higher fitting is selected to output the predicted value of the remaining effective life. The exponential decay model is as follows: ， The Weibull reliability model is as follows: ， in, The exponential decay coefficient is... For the Weibull scale parameter, For Weibull shape parameters.

9. The abnormal early warning method for the developer recovery system as described in claim 8, characterized in that: It also includes step S7: When any level of warning signal is triggered, a delayed shutdown request is sent. The delay time of the delayed shutdown request is the sum of the remaining time of the current batch and the fixed buffer time. If the warning signal disappears within the delay time, the delayed shutdown request is cancelled. If the warning signal continues until the end of the delay time, shutdown is performed.

10. The developer recovery system anomaly early warning method as described in claim 9, characterized in that: It also includes step S8: The warning events are stored in a non-volatile storage area. The warning events include the trigger time, the anomaly type, the concentration slope value at the time of triggering, and the concentration steady-state drift. When a warning reset command is received, the current warning status is cleared. If the root cause of the fault is not eliminated, the warning signal is triggered again within a preset time.

Images