Method for evaluating paint mist capture efficiency of a dry paint booth based on multi-sensor data

Abstract

The application discloses a kind of dry paint spray booth paint mist capture efficiency evaluation method based on multi-sensor data, it is related to monitoring and analysis technical field, this method includes: acquisition spraying event source signal, upstream spraying load, capture box before and after pressure difference signal and downstream concentration signal and constructs unified time axis;Identify candidate spraying event and set initial propagation window;Identify natural recovery section in stop spraying gap, sensor output is split into process attenuation component and self-pollution tail component, and corrected observation sequence and reliability coefficient are obtained by cross-validation of different sensors;Propagation time delay estimation value is obtained by limited cross-correlation search in initial propagation window;Accordingly, spraying cycle is segmented, and upstream spraying load and downstream escape amount are corresponded according to same physical batch, and capture efficiency is calculated.The application can improve the accuracy and stability of evaluation result.

Description

Technical Field

[0001] This application relates to the field of monitoring and analysis technology, and more specifically, to a method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data. Background Technology

[0002] Dry spray booths are widely used in automotive, construction machinery, and general industrial coating production lines. They capture paint mist generated during the spraying process through exhaust systems and paint mist separation devices to reduce downstream emissions and ensure stable equipment operation. In actual production, detection methods such as spray gun triggering, paint supply, pressure differential, particulate matter concentration, and volatile organic compound concentration are typically implemented to reflect the spraying load, separation device operating conditions, and downstream emissions. Current technologies for evaluating the paint mist capture efficiency of dry spray booths often involve directly comparing data from various sensors after simultaneous sampling, or constructing an efficiency index using upstream spraying load signals and downstream concentration signals at the same sampling time.

[0003] In conditions such as short-cycle spraying, rapid robot trajectory switching, color switching, and local saturation of drawer-type paint mist separation modules, the propagation time of the same batch of paint mist from the spraying area to the collection point and further transmission to downstream monitoring points is typically affected by factors such as the geometry of the spray booth, airflow, airflow organization, and duct routing. Particulate matter sensors and VOC sensors, when exposed to the paint mist environment for extended periods, are prone to adhesion contamination, response tailing, and baseline shift. If readings from different channels are directly spliced ​​or simply correlated using traditional methods, it is easy to misinterpret normal propagation delays as decreased collection efficiency and to misinterpret tailing caused by sensor contamination as the release of process residues, leading to distorted evaluation results that fail to accurately reflect the true collection status of the spray booth under different spraying cycles and operating stages. Summary of the Invention

[0004] This application provides a method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data, in order to at least solve some of the technical problems existing in the related technologies described above.

[0005] According to a first aspect of the embodiments of this application, a method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data is provided, including: Collect spraying event source signals, upstream spraying load, pressure difference signals before and after the collection box, and downstream concentration signals to construct a unified time axis; Candidate spraying events are identified based on the spraying event source signal, and an initial propagation window is set for each candidate spraying event according to the geometry of the spray booth and the air volume. During the interval between spray stops, the upper limit of the initial propagation window is used as the criterion for propagation completion, and the natural recovery segment is identified in combination with the system state criterion. Within the natural recovery phase, the output of each sensor is split into a process attenuation component and a self-contamination tail component. Through cross-validation of heterogeneous sensors, self-contamination and process residue release are distinguished, and a corrected observation sequence and a confidence coefficient are obtained. Within the initial propagation window, a restricted cross-correlation search is performed on the candidate spraying event trigger sequence and the correction observation sequence. The results of each channel are weighted by the confidence coefficient to obtain the propagation delay estimate. Based on the corrected observation sequence, the estimated propagation delay, and the confidence coefficient, each spraying cycle is divided into a spraying accumulation phase, a stable collection phase, an initial penetration phase, and a spraying stop recovery phase. Based on the estimated propagation delay, the upstream spraying load and downstream escape amount are causally correlated according to the same physical batch, and the acquisition efficiency is calculated based on the corrected observation sequence and the confidence coefficient.

[0006] As an optional approach, for the natural recovery segment, three criteria are simultaneously satisfied: the first criterion is that the spraying event source signal indicates that the spraying action has stopped; the second criterion is that the time elapsed since the end of the candidate spraying event has reached the upper limit of the corresponding initial propagation window; the third criterion is that the differential pressure signal shows a decreasing trend and the fan load tends to stabilize; when the attenuation trends of multiple heterogeneous sensors in the natural recovery segment are contradictory, the segment is marked as a low-confidence candidate segment and given a lower learning weight in the subsequent estimation of the self-contamination tail component.

[0007] As an alternative approach, the output of each sensor is decomposed into a process attenuation component and a self-contamination tail component. A low-order dynamic response model is used to parameterize the self-contamination tail characteristics of each sensor. The parameters of the low-order dynamic response model include the hysteresis time constant and the adhesion gain coefficient. Online identification is performed on the data in the natural recovery segment using the recursive least squares method. During the identification process, a forgetting factor is set to control the attenuation rate of historical data.

[0008] As an optional approach, the specific method for cross-validation of heterogeneous sensors is as follows: within the same natural recovery period, calculate the residual of each sensor's output relative to the physical attenuation constraint under the current ventilation conditions, and compare the consistency of the residuals of various sensors in terms of direction and amplitude; when the residuals of multiple heterogeneous sensors are consistent in direction and amplitude, attribute the corresponding attenuation deviation to the release of process residues; when only a few sensors deviate independently, attribute the attenuation deviation of that sensor to sensor self-contamination; the result of the consistency comparison is converted into the credibility coefficient of each sensor in the current time period.

[0009] As an optional approach, the reliability coefficient is between 0 and 1, and is determined by the self-contamination residual amplitude, residual repeatability, and consistency with heterogeneous sensors of the corresponding sensor in the most recent natural recovery periods. It is updated by an exponential weighted recursive method. For sensors whose confidence coefficient is higher than the preset lower threshold, a bounded inverse hysteresis correction method is used to compensate their output to generate the corrected observation sequence. The corrected observation sequence must meet the physical rationality constraints under the current airflow and pressure difference conditions. For sensors whose confidence coefficient is lower than the preset lower threshold, the participation weight of the sensor is reduced in the restricted cross-correlation search and the calculation of the acquisition efficiency.

[0010] As an optional approach, after obtaining the candidate delay values ​​from the constrained cross-correlation search, the candidate delay values ​​are progressively corrected using a constrained state-space estimation method, taking into account the changing trend of the differential pressure signal, the wind turbine operating condition, and the confidence coefficient, to obtain the propagation delay estimate. The observation noise variance of each sensor channel is inversely proportional to the confidence coefficient of that sensor, and when the confidence coefficient is lower than a preset lower threshold, it is processed according to a preset maximum observation noise variance.

[0011] As an optional approach, the initial penetration segment is divided based on the following: establishing a phase relationship and amplitude ratio relationship between the downstream concentration signal and the candidate spraying event after time delay alignment within the stable capture segment; when the phase relationship shows a continuous negative shift or the amplitude ratio relationship shows a continuous positive shift, and the shift cannot be explained by the concurrent air volume change and temperature and humidity fluctuation, the current time period is divided into the initial penetration segment.

[0012] As an optional approach, the calculation of the collection efficiency includes: within the stable collection section, estimating the effective collection ratio based on the upstream spraying load indicators, the change in the differential pressure signal, the current operating air volume, and the time-delayed aligned correction observation sequence; when multiple downstream sensors participate simultaneously, the contributions of each sensor are weighted according to the confidence coefficient. Within the initial penetration phase, the capture efficiency value at each time step is linearly fitted to obtain the efficiency decline rate; within the shutdown recovery phase, the decay curve of the corrected observation sequence is compared with the theoretical free decay curve under the current airflow conditions, and the cumulative area deviation between the two is output as the residual release index.

[0013] As an optional approach, the method further includes joint iteration: after completing the four-segment division of the spraying cycle, the natural recovery segment is re-examined based on the residual release index to see if it is affected by process residual release. If the residual release index is higher than a preset level, the learning weight of the corresponding natural recovery segment in the self-contamination tail component estimation is reduced. The parameters of the self-contaminated tail component are re-identified using the weighted natural recovery segment, and the confidence coefficient is updated. Then, the restricted cross-correlation search and four-segment division are re-executed using the updated corrected observation sequence and the confidence coefficient until the parameter changes of the self-contaminated tail component, the changes of the propagation delay estimate, and the segment boundary displacements between two adjacent iterations are all lower than their respective preset convergence thresholds.

[0014] According to a second aspect of the embodiments of this application, a system for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data is also provided, comprising: The acquisition module is used to acquire spraying event source signals, upstream spraying load, pressure difference signals before and after the collection box, and downstream concentration signals to construct a unified time axis; The event recognition and window setting module is used to identify candidate spraying events based on the spraying event source signal, and set an initial propagation window for each candidate spraying event according to the geometric dimensions and air volume of the spray booth; The natural recovery segment identification module is used to identify the natural recovery segment during the spray stop interval by using the upper limit of the initial propagation window as the propagation completion criterion and combining it with the system state criterion. The calibration processing module is used to split the output of each sensor into a process attenuation component and a self-contamination tail component during the natural recovery segment. The self-contamination and process residue release are distinguished by cross-validation of different types of sensors to obtain the calibration observation sequence and confidence coefficient. The propagation delay estimation module is used to perform a restricted cross-correlation search on the candidate spraying event trigger sequence and the correction observation sequence within the initial propagation window, and to obtain the propagation delay estimate by weighting the results of each channel with the confidence coefficient. The phase division module is used to divide each spraying cycle into a spraying accumulation phase, a stable collection phase, an initial penetration phase, and a spraying stop recovery phase based on the corrected observation sequence, the estimated propagation delay, and the confidence coefficient. The efficiency calculation module is used to correlate the upstream spraying load and the downstream escape amount with the same physical batch based on the propagation delay estimate, and to calculate the acquisition efficiency based on the corrected observation sequence and the confidence coefficient.

[0015] According to a third aspect of the embodiments of this application, an electronic device is provided, including: a processor; a memory for storing a computer program executable by the processor; wherein the processor is configured to execute the computer program in the memory to implement the method described in the first aspect.

[0016] According to a fourth aspect of the embodiments of this application, a computer-readable storage medium is provided, which, when an executable computer program in the storage medium is executed by a processor, enables the implementation of the method described in the first aspect.

[0017] This invention constructs a unified time axis based on the spraying event source signal, upstream spraying load, differential pressure signal, and downstream concentration signal. It also addresses the observation distortion caused by paint mist propagation delay and sensor self-contamination. By identifying the natural recovery phase, the sensor output is split into process attenuation and self-contamination tail components. Combined with cross-validation of dissimilar sensors and confidence coefficient weighting, reliable acquisition of the calibrated observation sequence is achieved. Based on this, a restricted cross-correlation search is used to determine the propagation delay, and the spraying cycle is divided into an initial spray accumulation phase, a stable collection phase, an initial penetration phase, and a shutdown recovery phase. This allows for causal correspondence calculation between the upstream spraying load and the downstream emission amount based on the same physical batch.

[0018] This not only reduces the misjudgment of normal propagation delay as efficiency anomaly and sensor tailing as process residue release, but also improves the accuracy and stability of dry spray booth paint mist collection efficiency assessment, and facilitates earlier identification of operating conditions such as local penetration and residue release.

[0019] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Furthermore, no embodiment in this disclosure is required to achieve all the effects described above. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0021] Figure 1 This is a schematic diagram of a method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data, provided in an embodiment of this disclosure.

[0022] Figure 2 This is a schematic diagram of the cross-validation process provided in an embodiment of this disclosure.

[0023] Figure 3 A schematic diagram illustrating the calculation process of the propagation delay estimate provided in the embodiments of this disclosure.

[0024] Figure 4This is a schematic diagram of a dry spray booth paint mist collection efficiency evaluation system based on multi-sensor data, provided in an embodiment of this disclosure.

[0025] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0027] According to embodiments of this disclosure, a method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data is provided. This method is applicable to dry spray booth systems equipped with drawer-type paint mist separation modules. The system deploys at least two of the following: differential pressure sensors, particulate matter concentration sensors, and volatile organic compound (VOC) concentration sensors. It also has at least one of the following as a spray gun trigger signal, robot spraying cycle signal, or paint supply valve switching signal as a spraying event source.

[0028] Under typical operating conditions such as short-cycle spraying, rapid robot trajectory switching, color switching, and drawer-type paint mist separation module tending to local saturation, there is a propagation time difference between the generation of the same batch of paint mist during spraying, the interception by the collection box, and the response at the downstream monitoring point. This time difference is determined by the air volume and pipeline direction. Furthermore, due to prolonged exposure to the paint mist adhering environment, each sensor gradually exhibits self-contamination phenomena such as response tailing and baseline shift.

[0029] When both propagation time delay and sensor self-contamination distortion exist simultaneously, the traditional method of directly stitching together sensor readings using synchronous sampling can easily misjudge normal propagation delay as an efficiency anomaly, and can easily misjudge the tailing caused by sensor contamination as the release of process residues, thus causing the evaluation results to deviate from the true situation. This method utilizes the aforementioned existing conventional sensor data to synchronously reconstruct the propagation time sequence of the same batch of paint mist and the reliable observation status of each sensor, achieving a stable evaluation of segmented collection efficiency.

[0030] The implementation process of the method described in this application will be described in detail below with reference to specific embodiments. It should be noted that this embodiment is only used to explain this application and is not intended to limit the scope of protection of this application. Conventional adjustments or substitutions of each step by those skilled in the art without departing from the concept of this application should be included in the scope of protection of this application.

[0031] Please see Figure 1 , Figure 1 This is a flowchart of a method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data, according to an embodiment of the present invention. Figure 1 As shown, the method includes steps S1-S7: In step S1, the spraying event source signal, upstream spraying load, pressure difference signal before and after the collection box, and downstream concentration signal are collected to construct a unified time axis.

[0032] In some embodiments, a spraying event source signal, an upstream spraying load characterization signal, a pressure difference signal across the collection box, and a downstream concentration signal are collected, and a unified time axis is constructed. The spraying event source signal includes at least one of a spray gun trigger signal, a robot spraying cycle signal, and a paint supply valve switch signal; the pressure difference signal across the collection box is measured by differential pressure sensors installed on the inlet and outlet sides of the collection box; the downstream concentration signal includes at least one of particulate matter concentration and total volatile organic compound concentration. The upstream spraying load characterization signal includes the duration of the spraying event, and the paint supply flow rate signal and / or spraying pressure signal collected when available. Optionally, auxiliary quantities such as exhaust fan frequency or power signals, temperature and humidity signals, and drawer in-position switch signals can also be connected.

[0033] Since the signals originate from different controllers, their sampling frequencies and timestamp bases often differ. Therefore, it is necessary to align all channels to the same time axis. The alignment method can employ a common industrial-grade unified time-base resampling approach: high-frequency signals are downsampled after anti-aliasing filtering, and low-frequency signals are filled using zero-order hold or linear interpolation. Simultaneously, missing data segments, frozen constant value segments, and communication interruption segments in each channel are marked. After the unified time axis is established, periods clearly unsuitable for efficiency evaluation, such as drawer pull-out and reset periods, equipment maintenance shutdown periods, and periods of abrupt fan start-up and shutdown transitions, are pre-marked. Subsequent processing automatically skips or downweights these periods; these pre-marks only serve as boundary protection.

[0034] In step S2, candidate spraying events are identified based on the spraying event source signal, and an initial propagation window is set for each candidate spraying event according to the geometry of the spray booth and the air volume.

[0035] Candidate spraying events are identified based on the spraying event source signal. The spray gun trigger signal is used as the priority event source, and the start and end times of each spraying action are extracted to form a sequence of candidate spraying events. Each candidate spraying event corresponds to a complete time period from the generation to the end of a batch of paint mist. In some embodiments, if the spray gun trigger channel is missing, the start and end times of the candidate spraying events are inferred by the approximately synchronous occurrence of three factors: a change in the state of the paint supply valve, a sharp rise in VOC concentration, and a disturbance in the differential pressure before the collection box. This inference is only used to anchor the time position and is not involved in the quantitative calculation of efficiency values.

[0036] For each candidate spraying event, an initial propagation window is set based on the spray booth geometry and airflow. The initial propagation window is a time interval used to define the range of propagation time required for the same batch of paint mist to travel from the spraying area to each downstream monitoring location. The lower limit of the window corresponds to the theoretical shortest propagation time for paint mist to reach the downstream monitoring point along the main airflow direction under the current airflow conditions. This value mainly depends on the straight-line distance between the spraying area and the monitoring point and the current mainstream wind speed. The upper limit of the window corresponds to the theoretical longest propagation time for paint mist to reach the same monitoring point in low-wind-speed areas or on paths with backflow, while also allowing for a margin to cover deviations caused by airflow fluctuations.

[0037] The initial propagation window defines the physical search boundary for propagation delay inversion in subsequent processes, and provides the time range within which the response of the paint mist process may still exist for the identification of the natural recovery segment, thereby avoiding the erroneous treatment of the normal response of the paint mist still in the propagation process as sensor tailing.

[0038] In some embodiments, for step S3, the upper limit of the initial propagation window is used as the propagation completion criterion during the spray stop interval, and the natural recovery segment is identified in combination with the system state criterion.

[0039] During the intervals between adjacent candidate spraying events, the upper limit of the initial propagation window is used as the criterion for propagation completion, combined with system state criteria, to identify the natural recovery phase. The natural recovery phase refers to the period when the spraying source has stopped, and the paint mist has essentially completed its propagation from the spraying area to the main monitoring locations, and the system is in a state of concentration decay and depletion. In this method, it serves as a natural reference window for the sensor's online self-contamination learning.

[0040] In some embodiments, the identification of the natural recovery segment requires the simultaneous satisfaction of three criteria. The first criterion is that the spraying event source signal indicates that the spraying action has stopped, i.e., the spray gun is closed or the paint valve is shut off. The second criterion focuses on whether the propagation is complete: the time elapsed since the end of the candidate spraying event must reach the upper limit of the corresponding initial propagation window to ensure that the batch of paint mist has essentially reached each monitoring location through the main propagation path under the current airflow conditions. This criterion is introduced because if the time elapsed after spraying stops is insufficient for the paint mist to propagate downstream, then the concentration response detected by the downstream monitoring point at this time is part of the normal propagation process and is not a false signal caused by sensor self-contamination; only the residual response that persists after the propagation window time limit has been reached is likely to be the sensor's own tailing behavior.

[0041] The third criterion examines the system status: the differential pressure signal before and after the collection box shows a decreasing trend, the exhaust fan load tends to stabilize, and the temperature and humidity do not fluctuate drastically. All of the above indicators indicate that the system is entering the normal exhaust attenuation stage.

[0042] In practice, the spraying event source signal is used to locate the time of spraying stop. Then, the start time of the candidate recovery segment is calculated based on the current air volume and the corresponding upper limit of the initial propagation window. Finally, the effectiveness of the candidate recovery segment is confirmed by combining the pressure difference decrease trend, fan load changes and whether each downstream channel has entered the attenuation phase.

[0043] The confirmation method can employ change point detection based on physical exhaust attenuation characteristics. Specifically, when signals from multiple monitoring channels all exhibit a transition from fluctuation to monotonic attenuation within the same time window, the change point is the starting point of the recovery segment. The role of change point detection here is to automatically locate moments in continuous data where significant changes in signal statistical characteristics occur. A common implementation method is to compare the signal's mean or variance using a sliding window; when the difference between adjacent windows exceeds a preset threshold, it is considered a change point.

[0044] According to the embodiments of this disclosure, the input for change point detection is the signal segment of each channel within the candidate interval of the recovery segment after being aligned with a unified time axis. The detection target is the turning point where the signal changes from high fluctuations during spraying to monotonically decaying after spraying stops. The consistency of the change point positions of multiple channels further enhances the reliability of the determination.

[0045] If, within a candidate recovery segment, the attenuation trends of various physical quantities from sensors, such as particulate matter concentration, VOC concentration, and differential pressure, show significant contradictions, then this segment will not be marked as a high-confidence recovery segment. Instead, it will be temporarily retained as a low-confidence candidate segment and assigned a lower learning weight in the subsequent estimation of the self-contamination tail component, in order to reduce the interference of uncertain recovery segments on parameter identification. The search starting point for change point detection is no earlier than the time corresponding to the upper limit of the initial propagation window; when the detected turning point time is later than the time corresponding to the upper limit of the initial propagation window, the later time is used as the starting point of the recovery segment.

[0046] In step S4, during the natural recovery phase, the outputs of each sensor are split into process attenuation components and self-contamination tail components. Self-contamination and process residue release are distinguished through cross-validation of different types of sensors, and the corrected observation sequence and confidence coefficient are obtained.

[0047] Please see Figure 2 , Figure 2 A schematic diagram of the cross-validation process provided in an embodiment of this disclosure is shown. For example... Figure 2 As shown, in step S201, the process attenuation component and the self-contamination tail component are separated.

[0048] The actual paint mist concentration within the recovery phase should exhibit a monotonically decreasing or segmented decreasing trend under current ventilation conditions. This is determined by the physical processes of spatial dilution and the removal of paint mist particles by exhaust air. Using this as a constraint, if the output of a sensor exhibits a slow decline, a prolonged tail, or an overall upward shift in the baseline during the expected decay period, and this deviation occurs repeatedly in multiple recovery phases, it can be concluded that the sensor's own deposits or residual sampling channels have caused the tailing component. Specifically, the sensor's dynamic response pattern within the recovery phase is compared with the current system's physical venting process, focusing on whether the pattern conforms to the expected decay rather than whether the numerical value exceeds the limit.

[0049] In step S202, the self-contamination tailing characteristics of each sensor are parameterized using a low-order dynamic response model. This model represents the actual output of the sensor as the result of the real process signal after passing through the sensor's own response characteristics. The model parameters include two items: hysteresis time constant and adhesion gain coefficient. The hysteresis time constant characterizes the response delay caused by the adhesion layer on the sensor surface, and the adhesion gain coefficient describes the degree to which the adhesion layer attenuates or amplifies the sensitivity.

[0050] In some embodiments, a first-order hysteresis model is selected, which requires only one set of time constants and gain coefficients to characterize the tailing behavior. When the sensor exhibits deviations at two different time scales—fast tailing and slow baseline rise—a second-order hysteresis model can be used, adding a second set of parameters to describe the fast and slow components separately. Optionally, a first-order model is preferred when the recovery period is short or the signal-to-noise ratio is low to avoid instability in parameter identification due to insufficient data. When the recovery period is long enough and the sensor tailing pattern indeed exhibits the characteristics of two superimposed attenuation rates, a second-order model is used instead.

[0051] The parameters of the low-order dynamic response model are identified online on the natural recovery segment data using the recursive least squares (RLS) method. Recursive least squares is a classic algorithm that incrementally updates parameter estimates whenever new data arrives. The calculation process is as follows: when each recovery segment arrives, an error signal is constructed using the deviation sequence between the sensor output and the physical attenuation constraint. This error signal drives the parameter estimates to be updated in the direction of reducing the deviation, without needing to store all historical recovery segment data.

[0052] During the identification process, a forgetting factor can be set to control the decay rate of historical data. The forgetting factor can be configured between 0.90 and 0.99. The smaller the value, the more emphasis is placed on the identification results of the recent recovery period and the more sensitive the tracking of parameter changes. The larger the value, the smoother the parameter estimation. For example, the forgetting factor can be set to 0.95.

[0053] In step S203, cross-validation with different types of sensors is used to distinguish between self-contamination and process residue release, and the corrected observation sequence and confidence coefficient are obtained.

[0054] Judging tailing solely by the attenuation pattern of a single sensor type is insufficient if the trapping medium itself is still slowly releasing paint mist within the recovery section. In such cases, deviations from attenuation constraints in sensor outputs may not necessarily be due to self-contamination. To differentiate between sensor self-contamination and process residue release, cross-validation of dissimilar sensors is performed within the same natural recovery section.

[0055] Specifically, the residuals of each sensor's output relative to the physical attenuation constraint under the current ventilation conditions are calculated, i.e., the deviation between the measured attenuation curve and the theoretical attenuation curve. Then, the residuals of different types of sensors are compared for consistency in direction and amplitude. When the residuals of multiple heterogeneous sensors show consistency in direction and amplitude, for example, when both particulate matter concentration and VOC concentration are synchronously high and the pressure difference also indicates that there is still slow release in the collection box, the corresponding attenuation deviation is attributed to the release of process residues.

[0056] When only a few individual sensors exhibit independent deviations, such as severe tailing of particulate matter concentration while VOC concentration and differential pressure have returned to normal during the same period, the attenuation deviation of that sensor is attributed to sensor self-contamination. The results of the consistency comparison are ultimately converted into the reliability coefficient of each sensor in the current time period.

[0057] The reliability coefficient ranges from 0 to 1, and is determined by a combination of the self-contamination residual magnitude, residual repeatability, and consistency with dissimilar sensors in the most recent natural recovery segments. The determination method employs an exponentially weighted recursive update: whenever a new natural recovery segment completes cross-validation, the reliability coefficient of the sensor is updated using the consistency comparison results of that segment with an exponentially decaying weight, ensuring that the evaluation results of recent recovery segments have a greater impact on reliability than earlier recovery segments.

[0058] For example, if the residual direction of a certain sensor is consistent with that of a different type of sensor in multiple consecutive recovery segments, its confidence coefficient tends to be 1; if it exhibits an independent and gradually increasing deviation pattern in the consecutive recovery segments, its confidence coefficient gradually decreases.

[0059] In step S204, based on the estimation results and confidence coefficients of the self-contamination tail component, a calibration observation sequence is generated for each sensor.

[0060] For sensors with a confidence coefficient higher than a preset lower threshold, a bounded inverse hysteresis correction method is used to compensate their output and generate a corrected observation sequence. Bounded inverse hysteresis correction means using the identified low-order dynamic response model parameters to perform inverse compensation on the sensor output to restore the true process signal, but the compensation result must meet the physical rationality constraints under the current air volume and pressure difference conditions.

[0061] The specific compensation process involves performing an inverse filtering operation on the sensor output sequence based on the identified hysteresis time constant and attached gain coefficient. This removes the tailing component caused by the sensor response delay, resulting in a corrected sequence that approximates the actual process signal. Inverse filtering is the reverse operation of low-pass filtering; its essence is to amplify the high-frequency components in the signal to compensate for the broadening caused by the filter. However, if this amplification is not restricted, it can easily introduce numerical oscillations or even produce physically unreasonable extreme values.

[0062] The physical rationality constraints include ensuring that the corrected concentration sequence does not exhibit negative values ​​at any time, does not show unfounded spikes exceeding the upper limit corresponding to the current spraying load, and does not exhibit violent oscillations contrary to the trends of pressure difference and airflow changes. When the inverse filtering result touches the above boundaries at a certain moment, the value at that moment is truncated to the boundary value to prevent overcorrection.

[0063] For sensors with a confidence coefficient below a preset lower threshold, i.e., when contamination is so severe that the low-order model is unable to correct it, inverse hysteresis correction is not forcibly performed. Instead, the participation weight of that sensor channel is reduced in subsequent restricted cross-correlation search and acquisition efficiency calculations. In some embodiments, the calibrated observations of other high-confidence sensors and the current system operating state can be used to infer alternative estimates of the physical quantities corresponding to the low-confidence sensors through physical correlation, allowing these alternative values ​​to participate in subsequent calculations.

[0064] In step S5, within the initial propagation window, a restricted cross-correlation search is performed on the candidate spraying event trigger sequence and the correction observation sequence. The results of each channel are weighted by the confidence coefficient to obtain the propagation delay estimate.

[0065] Please see Figure 3 , Figure 3 A schematic diagram illustrating the calculation process for the propagation delay estimate provided in this embodiment of the disclosure. Figure 3 As shown, in step S301, a restricted cross-correlation search is performed on the candidate spraying event trigger sequence and the correction observation sequence within the initial propagation window.

[0066] The specific process of the restricted cross-correlation search is as follows: for each candidate spraying event, its trigger sequence is used as the reference signal, and the corrected response sequences of each downstream channel are used as the search signals. The cross-correlation function values ​​of the two are calculated point by point within the time delay search range defined by the initial propagation window, and the time delay position corresponding to the peak value of the cross-correlation function is taken as the candidate time delay value of the channel.

[0067] Cross-correlation is a classic time delay estimation method in signal processing. Its basic principle is to measure the similarity between two signals at different relative offsets, and the offset corresponding to the highest similarity is the time delay between the two signals. Since the downstream response sequences involved in the search have undergone self-contamination correction and confidence redistribution, the spurious cross-correlation peaks caused by sensor tailing are effectively suppressed; in addition, the search range is limited to within the initial propagation window, and the time delay estimation will not deviate from the physical propagation range allowed by the current system conditions.

[0068] In step S302, after obtaining the candidate time delay values ​​for each channel, the propagation time delay is finally obtained by combining the changing trend of the differential pressure signal, the operating condition of the fan, and the confidence coefficient with a constrained state-space estimation method for segment-by-segment correction.

[0069] In some embodiments, state-space estimation is implemented using a Kalman filter with boundary conditions, where the state variable is the propagation delay at each monitoring location. The Kalman filter is a recursive state estimation algorithm that updates the state estimate at each time step based on a weighted fusion of predicted and observed values.

[0070] Here, the process noise variance reflects the amplitude of the disturbance caused by airflow fluctuations to the propagation time delay. The observation noise variance is inversely proportional to the confidence coefficient of each sensor channel; that is, the higher the confidence level of a channel, the greater its contribution to the time delay estimation through its cross-correlation peak, while channels with lower confidence levels are automatically weighted less. When the confidence coefficient is lower than a preset lower threshold, the channel is treated according to the preset maximum observation noise variance. In other embodiments, time delay correction can also be achieved by using change point detection combined with restricted cross-correlation, or by using an equivalent recursive least squares time delay estimation method.

[0071] Thus, a mutually supportive relationship is formed between propagation delay inversion and sensor self-contamination correction. Self-contamination correction improves the reliability of the downstream response sequence, making it easier for cross-correlation search to locate the true propagation delay. The initial propagation window also defines the physical boundary for recovery segment identification, ensuring that recovery segment learning does not incorrectly include signals that are still propagating normally.

[0072] In step S6, based on the corrected observation sequence, the estimated propagation delay, and the confidence coefficient, each spraying cycle is divided into a spraying start-up accumulation stage, a stable collection stage, an initial penetration stage, and a spraying stop recovery stage.

[0073] Based on the calibrated observation sequence, propagation delay estimate, and confidence coefficient, each spraying cycle is divided into four stages: initial spray accumulation stage, stable collection stage, initial penetration stage, and shutdown recovery stage. All inputs used for this division are calibrated observation sequences that have undergone self-contamination correction and confidence redistribution.

[0074] The initial spray accumulation phase covers the period from the start of spraying until the downstream response begins to stabilize after time-delay alignment. Based on current propagation delay estimates, the concentration response at downstream monitoring points is still rising during this phase, and efficiency assessment lacks complete upstream-downstream comparison conditions. Therefore, the initial spray accumulation phase is used to determine the boundary of the early stage of the spraying cycle and is not considered an independent capture efficiency output phase.

[0075] The stable collection stage refers to the stage in which the pressure difference before and after the collection box and the downstream particulate matter concentration and VOC concentration after time delay alignment exhibit a stable interception relationship. At this time, the downstream escape amount after the upstream spraying load is intercepted by the collection box remains at a stable level.

[0076] The initial penetration section is divided based on establishing the phase relationship and amplitude ratio between the downstream concentration signal and the candidate spraying event after time delay alignment within the stable capture section. The former measures whether the time delay offset of the downstream response relative to the spraying event is stable, and the latter measures whether the ratio of the downstream escaping concentration to the upstream spraying load is stable. These two relationships are then continuously monitored.

[0077] When the phase relationship shows a sustained negative shift, that is, the propagation delay is abnormally shortened, it means that the paint mist bypasses the saturated area in the collection medium and penetrates to the downstream earlier along a shorter path; or when the amplitude ratio shows a sustained positive shift, that is, the escape ratio is abnormally increased; and at the same time, the above shifts cannot be explained by the concurrent air volume changes and temperature and humidity fluctuations, then the current period is divided into the initial penetration segment.

[0078] The method for determining an inexplicable shift is to extract the regression components corresponding to changes in airflow and temperature / humidity from the changes in phase shift and amplitude ratio shift, respectively, and then calculate the residuals. If the residuals exceed the statistical fluctuation range established during the stable acquisition period, the shift is considered abnormal. The regression components are obtained through a regression model with airflow and temperature / humidity as independent variables and phase shift and amplitude ratio shift as dependent variables.

[0079] This phase-relationship-based penetration criterion, at the physical level, utilizes the shortening path of paint mist particles when the trapping medium begins to saturate locally or forms structural penetration channels. Compared to the traditional approach that relies on the downstream absolute concentration exceeding a fixed threshold, it can detect early local failure trends even when the outlet concentration is still within acceptable limits. Since the calibrated observation sequence has eliminated spurious phase shifts caused by sensor tailing, the stability of this criterion is correspondingly improved during long-term operation in continuous production.

[0080] The spray-stop recovery phase corresponds to the stage after spraying stops, during which the system enters a period of evacuation and decay. It also serves as the training period for the next round of sensor self-contamination learning. The evaluation result for this phase is the residual release index.

[0081] In step S7, the upstream spraying load and downstream escape amount are causally correlated according to the same physical batch based on the propagation delay estimate, and the capture efficiency is calculated based on the corrected observation sequence and the confidence coefficient.

[0082] Based on the estimated propagation delay, the upstream spraying load and downstream escaping volume are causally correlated according to the same physical batch. The collection efficiency of each segment is calculated based on the calibrated observation sequence and confidence coefficient. In the traditional approach, the efficiency is estimated by directly taking the ratio of upstream and downstream sensor readings at the same sampling time. This is equivalent to assuming that the paint mist occurs instantaneously from generation to detection. In reality, under the geometric dimensions and airflow conditions of a dry spray booth, the propagation delay can reach several seconds or even more than ten seconds.

[0083] By shifting the upstream load signal of each candidate spraying event backward according to the propagation delay estimate, aligning it with the corresponding downstream corrected observation sequence on the time axis to the same physical batch, the established comparison relationship is used to reflect the escape of the same batch of paint mist after being intercepted by the collection box.

[0084] Within the stable capture phase, the upstream spraying load indicator is determined by the duration of the spraying event and corrected by combining the paint supply flow rate signal and / or spraying pressure signal for the corresponding time period when available. The downstream effluent indicator is characterized by the product of the corrected and time-delay-aligned particulate matter concentration or VOC concentration and the current airflow. The capture efficiency is estimated based on a comparison between the two indicators mentioned above. When multiple downstream sensors participate simultaneously, the contribution of each sensor to the effluent estimation is weighted according to its respective confidence coefficient, with higher confidence sensor channels having larger weights and lower confidence channels having smaller weights.

[0085] Within the initial penetration segment, in addition to outputting the capture efficiency value for that segment, a linear fit is performed on the capture efficiency values ​​at each time step within the initial penetration segment to obtain the efficiency decrease rate. This rate reflects the speed at which the drawer-type paint mist separation module transitions from a stable interception state to local penetration. Optionally, if the penetration trend exhibits nonlinear characteristics, further piecewise fitting can be performed to describe the accelerated penetration process.

[0086] During the recovery phase after the spraying stop, the decay curve of the calibrated observation sequence is compared with the theoretical free decay curve under the current airflow conditions. The theoretical free decay curve is calculated based on the concentration level at the beginning of the recovery phase and the current ventilation conditions, and its decay rate depends on the ratio of the effective volume of the spray booth to the exhaust volume.

[0087] The cumulative area deviation between the two during the recovery period is output as the residual release index. A residual release index close to zero indicates that the paint mist concentration has dissipated normally after spraying stops, while a significantly positive index indicates that the collecting medium or the inner wall of the pipeline is still slowly releasing paint mist. This index has both process evaluation significance and sensor calibration auxiliary significance. When the residual release index shows consistent performance on multiple types of sensors, it further supports attributing the corresponding attenuation deviation to process residual release rather than sensor self-contamination.

[0088] Optionally, when the overall confidence coefficient of each sensor is lower than the preset level within a certain spraying cycle, the capture efficiency output of that cycle is marked as a low confidence result and accompanied by a confidence level, instead of forcibly giving a highly deterministic efficiency value.

[0089] In some embodiments, the method further includes joint iteration and convergence. The estimation of the self-contaminated tail component, the update of the confidence coefficient, the propagation delay inversion, and the four-segment partitioning are interdependent. The identification range of the recovery segment is constrained by the initial propagation window; the parameter identification of the self-contaminated tail component is completed within the recovery segment; the corrected observation sequence and confidence coefficient affect the accuracy of the restricted cross-correlation search; and the result of the four-segment partitioning, in turn, affects the judgment of the quality of the recovery segment.

[0090] To ensure collaborative work among all stages, joint iterations can be performed in the following order within each analysis window: First, locate candidate recovery segments and identify initial self-contamination parameters to obtain initial calibration observation sequences and initial confidence coefficients. Then, based on these initial results, complete restricted cross-correlation search and four-segment division. Next, re-examine whether the natural recovery segment is affected by process residue release based on the residual release index in the division results. If the residual release index is higher than the preset level, it indicates that the trapping medium is still releasing paint mist during the time period corresponding to the start time of the recovery segment, and process components are mixed into the recovery segment data. At this time, reduce the learning weight of the corresponding natural recovery segment in the self-contamination tail component estimation to reduce the interference of process residue release on sensor calibration parameters. Subsequently, the parameters of the self-contaminated tail component are re-identified using the weighted recovery segment, and the confidence coefficient is updated. Then, the restricted cross-correlation search and four-segment division are re-executed using the updated corrected observation sequence and confidence coefficient. This process is repeated until the changes in the parameters of the self-contaminated tail component, the changes in the propagation delay estimate, and the segment boundary displacement between adjacent iterations are all lower than their respective preset convergence thresholds.

[0091] The preset convergence threshold depends on the engineering accuracy requirements and can be set independently for the tail parameter, delay, and segment boundary. Specifically, the convergence threshold for the tail parameter is set according to the parameter estimation resolution, while the convergence thresholds for propagation delay and segment boundary are set according to the sampling interval of a unified time axis. Since the search range for propagation delay is physically limited, and the parameter space dimension of the low-order response model is finite, iterations typically converge within a relatively small number of iterations. For example, the upper limit of the number of iterations can be configured to three to five.

[0092] When the results of two adjacent iterations in the joint iteration step oscillate and fail to meet the preset convergence threshold, the weighted average of the results of the two most recent iterations is taken as the output of the spraying cycle, and the corresponding capture efficiency is marked as a low confidence result to indicate to subsequent users that the evaluation uncertainty of the cycle is high.

[0093] In some embodiments, during continuous production, when a drawer replacement event, a step change in the fan frequency setpoint, or a sensor restart after cleaning is detected, the system's historical learning parameters should not be continued. These events are collectively referred to as structural change points, and their typical detection criteria include the flipping of the drawer in-position switch state, the jump of the maintenance completion signal, and a step change in the fan frequency setpoint.

[0094] Once a structural change is confirmed, key parameters need to be reset or rapidly updated. Specifically, the parameters of the low-order dynamic response model are reset to their initial default values, the initial propagation window is reset to the theoretical range calculated based on the current airflow and geometry, and historical reliability assessment records for the natural recovery phase are cleared. The initial default values ​​are the calibration values ​​after equipment commissioning or sensor cleaning.

[0095] After resetting, the forgetting factor is temporarily lowered, for example, from the normal range to near its lower limit, so that the system responds faster to newly arrived recovery segment data after the change point and the parameter updates are more sensitive. After accumulating enough natural recovery segments under the new operating conditions to complete parameter relearning, the forgetting factor is gradually restored to the normal range. The number of sufficient natural recovery segments is preset according to the number of parameters in the low-order dynamic response model. The above processing avoids the continued use of sensor calibration parameters and propagation delay priors from the old operating conditions after drawer replacement or media update, thereby maintaining the long-term adaptability of the evaluation system under cross-seasonal, cross-coating system, and cross-drawer state conditions.

[0096] This method combines the recovery of the propagation time sequence of paint mist from the same batch with online self-contamination correction of the sensor in the same processing chain, changing the traditional approach of processing time delay correction and sensor correction independently. The physical boundary of the propagation window constrains the effective range of the recovered segment identification, ensuring that the sensor correction learns the true self-contamination tail rather than process propagation residue; the corrected observation sequence and confidence coefficient, in turn, improve the accuracy of propagation time delay inversion.

[0097] The segmented efficiency calculation is based on the upstream and downstream causal correspondence of the same physical batch after time delay alignment. Combined with the phase relationship criterion of the initial penetration segment and the residual release index of the recovery segment, the evaluation results remain stable under complex working conditions such as short-cycle spraying, fan frequency conversion adjustment and the initial stage of local saturation of drawer-type paint mist separation module. Furthermore, it can maintain reliable efficiency output even when the sensor gradually becomes dirty after long-term operation.

[0098] Please see Figure 4 , Figure 4 This is a schematic diagram of a dry spray booth paint mist collection efficiency evaluation system based on multi-sensor data, provided in an embodiment of this application. Figure 4 As shown, the system includes: The acquisition module 401 is used to acquire spraying event source signals, upstream spraying load, pressure difference signals before and after the collection box, and downstream concentration signals to construct a unified time axis; The event recognition and window setting module 402 is used to identify candidate spraying events based on the spraying event source signal, and set an initial propagation window for each candidate spraying event based on the geometric dimensions and air volume of the spray booth; The natural recovery segment identification module 403 is used to identify the natural recovery segment during the spray stop interval by using the upper limit of the initial propagation window as the propagation completion criterion and combining it with the system state criterion. The calibration processing module 404 is used to split the output of each sensor into a process attenuation component and a self-contamination tail component during the natural recovery segment, and to distinguish between self-contamination and process residue release through cross-validation of heterogeneous sensors, thereby obtaining a calibration observation sequence and a confidence coefficient. The propagation delay estimation module 405 is used to perform a restricted cross-correlation search on the candidate spraying event trigger sequence and the correction observation sequence within the initial propagation window, and to obtain the propagation delay estimate by weighting the results of each channel with the confidence coefficient. The stage division module 406 is used to divide each spraying cycle into a spraying accumulation stage, a stable collection stage, an initial penetration stage, and a spraying stop recovery stage based on the corrected observation sequence, the propagation delay estimate, and the confidence coefficient. The efficiency calculation module 407 is used to correspond the upstream spraying load and the downstream escape amount to the same physical batch based on the propagation delay estimate, and to calculate the acquisition efficiency based on the corrected observation sequence and the confidence coefficient.

[0099] Each processing unit and / or module in the embodiments of this application can be implemented by an analog circuit that implements the functions described in the embodiments of this application, or by software that executes the functions described in the embodiments of this application.

[0100] Based on the same inventive concept, this application also provides an electronic device, the method corresponding to which can be the method in the foregoing embodiments, and its problem-solving principle is similar to that method. For example... Figure 5 As shown, Figure 5 This is a schematic diagram of an electronic device structure provided in an embodiment of the present disclosure. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the methods and / or technical solutions of the foregoing embodiments of the present application.

[0101] In particular, the methods and / or embodiments in this application can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. When the computer program is executed by a processor, it performs the functions defined in the methods of this application.

[0102] Another embodiment of this application provides a storage medium storing computer program instructions thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application.

[0103] In the above embodiments, the descriptions of each embodiment have different focuses. Parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. The above descriptions are merely preferred embodiments of this application and explanations of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by specific combinations of the above technical features, but should also cover other technical solutions formed by arbitrary combinations of the above technical features or their equivalent features without departing from the inventive concept.

Claims

1. A method for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data, characterized in that, include: Collect spraying event source signals, upstream spraying load, pressure difference signals before and after the collection box, and downstream concentration signals to construct a unified time axis; Candidate spraying events are identified based on the spraying event source signal, and an initial propagation window is set for each candidate spraying event according to the geometry of the spray booth and the air volume. During the interval between spray stops, the upper limit of the initial propagation window is used as the criterion for propagation completion, and the natural recovery segment is identified in combination with the system state criterion. Within the natural recovery phase, the output of each sensor is split into a process attenuation component and a self-contamination tail component. Through cross-validation of heterogeneous sensors, self-contamination and process residue release are distinguished, and a corrected observation sequence and a confidence coefficient are obtained. Within the initial propagation window, a restricted cross-correlation search is performed on the candidate spraying event trigger sequence and the correction observation sequence. The results of each channel are weighted by the confidence coefficient to obtain the propagation delay estimate. Based on the corrected observation sequence, the estimated propagation delay, and the confidence coefficient, each spraying cycle is divided into an initial spray accumulation phase, a stable collection phase, an initial penetration phase, and a stop-spray recovery phase. Based on the estimated propagation delay, the upstream spraying load and downstream escape amount are causally correlated according to the same physical batch, and the acquisition efficiency is calculated based on the corrected observation sequence and the confidence coefficient.

2. The method according to claim 1, characterized in that, For the natural recovery segment, three criteria are simultaneously satisfied: the first criterion is that the spraying event source signal indicates that the spraying action has stopped; the second criterion is that the time elapsed since the end of the candidate spraying event has reached the upper limit of the corresponding initial propagation window; the third criterion is that the differential pressure signal shows a decreasing trend and the fan load tends to stabilize; when the attenuation trends of multiple heterogeneous sensors in the natural recovery segment are contradictory, the segment is marked as a low-confidence candidate segment and given a lower learning weight in the subsequent estimation of the self-contamination tail component.

3. The method according to claim 1, characterized in that, The outputs of each sensor are decomposed into process attenuation components and self-contamination tail components. A low-order dynamic response model is used to parameterize the self-contamination tail characteristics of each sensor. The parameters of the low-order dynamic response model include the hysteresis time constant and the adhesion gain coefficient. Online identification is performed on the data of the natural recovery segment using the recursive least squares method. During the identification process, a forgetting factor is set to control the attenuation rate of historical data.

4. The method according to claim 1, characterized in that, The specific method for cross-validation of heterogeneous sensors is as follows: within the same natural recovery period, the residuals of each sensor's output relative to the physical attenuation constraints under the current ventilation conditions are calculated, and the residuals of various types of sensors are compared in terms of direction and amplitude. When the residuals of multiple heterogeneous sensors are consistent in direction and amplitude, the corresponding attenuation deviation is attributed to the release of process residues. When only a few sensors deviate independently, the attenuation deviation of that sensor is attributed to sensor self-contamination. The result of the consistency comparison is converted into the credibility coefficient of each sensor in the current time period.

5. The method according to claim 1, characterized in that, The reliability coefficient ranges from 0 to 1, and is determined by the self-contamination residual amplitude, residual repeatability, and consistency with heterogeneous sensors in the most recent natural recovery periods of the corresponding sensor. It is updated by an exponential weighted recursive method. For sensors whose confidence coefficient is higher than the preset lower threshold, a bounded inverse hysteresis correction method is used to compensate their output to generate the corrected observation sequence. The corrected observation sequence must meet the physical rationality constraints under the current airflow and pressure difference conditions. For sensors whose confidence coefficient is lower than the preset lower threshold, the participation weight of the sensor is reduced in the restricted cross-correlation search and the calculation of the acquisition efficiency.

6. The method according to claim 1, characterized in that, After obtaining the candidate time delay values ​​from the constrained cross-correlation search, the candidate time delay values ​​are progressively corrected using a constrained state-space estimation method, taking into account the changing trend of the differential pressure signal, the wind turbine operating condition, and the confidence coefficient, to obtain the propagation time delay estimate. The observation noise variance of each sensor channel is inversely proportional to the confidence coefficient of that sensor, and when the confidence coefficient is lower than a preset lower threshold, it is processed according to the preset maximum observation noise variance.

7. The method according to claim 1, characterized in that, The initial penetration segment is defined as follows: within the stable capture segment, a phase relationship and amplitude ratio relationship are established between the downstream concentration signal after time delay alignment and the candidate spraying event. When the phase relationship shows a continuous negative shift or the amplitude ratio relationship shows a continuous positive shift, and the shift cannot be explained by the concurrent air volume change and temperature and humidity fluctuation, the current time period is defined as the initial penetration segment.

8. The method according to claim 1, characterized in that, The calculation of the collection efficiency includes: within the stable collection section, estimating the effective collection ratio based on the upstream spraying load indicator, the change of the differential pressure signal, the current operating air volume, and the time-delayed aligned correction observation sequence; when multiple downstream sensors participate simultaneously, the contributions of each sensor are weighted according to the confidence coefficient. Within the initial penetration phase, the capture efficiency value at each time step is linearly fitted to obtain the efficiency decline rate; within the shutdown recovery phase, the decay curve of the corrected observation sequence is compared with the theoretical free decay curve under the current airflow conditions, and the cumulative area deviation between the two is output as the residual release index.

9. The method according to claim 8, characterized in that, The method also includes joint iteration: after completing the four-segment division of the spraying cycle, the natural recovery segment is re-examined based on the residual release index to see if it is affected by process residual release. If the residual release index is higher than the preset level, the learning weight of the corresponding natural recovery segment in the self-contamination tail component estimation is reduced. The parameters of the self-contaminated tail component are re-identified using the weighted natural recovery segment, and the confidence coefficient is updated. Then, the restricted cross-correlation search and four-segment division are re-executed using the updated corrected observation sequence and the confidence coefficient until the parameter changes of the self-contaminated tail component, the changes of the propagation delay estimate, and the segment boundary displacements between two adjacent iterations are all lower than their respective preset convergence thresholds.

10. A system for evaluating the paint mist collection efficiency of a dry spray booth based on multi-sensor data, characterized in that, include: The acquisition module is used to acquire spraying event source signals, upstream spraying load, pressure difference signals before and after the collection box, and downstream concentration signals to construct a unified time axis; The event recognition and window setting module is used to identify candidate spraying events based on the spraying event source signal, and set an initial propagation window for each candidate spraying event according to the geometric dimensions and air volume of the spray booth; The natural recovery segment identification module is used to identify the natural recovery segment during the spray stop interval by using the upper limit of the initial propagation window as the propagation completion criterion and combining it with the system state criterion. The calibration processing module is used to split the output of each sensor into a process attenuation component and a self-contamination tail component during the natural recovery segment. The self-contamination and process residue release are distinguished by cross-validation of different types of sensors to obtain the calibration observation sequence and confidence coefficient. The propagation delay estimation module is used to perform a restricted cross-correlation search on the candidate spraying event trigger sequence and the correction observation sequence within the initial propagation window, and to obtain the propagation delay estimate by weighting the results of each channel with the confidence coefficient. The phase division module is used to divide each spraying cycle into a spraying accumulation phase, a stable collection phase, an initial penetration phase, and a spraying stop recovery phase based on the corrected observation sequence, the estimated propagation delay, and the confidence coefficient. The efficiency calculation module is used to correlate the upstream spraying load and the downstream escape amount with the same physical batch based on the propagation delay estimate, and to calculate the acquisition efficiency based on the corrected observation sequence and the confidence coefficient.

Images