Flow sensor data processing method and system based on multi-dimensional compensation model

Abstract

The invention relates to the technical field of flow sensor data processing, and discloses a flow sensor data processing method and system based on a multi-dimensional compensation model.The method comprises the steps of extracting a process main period and generating a temperature drift period template, detecting a medium switching event and generating phase-switching probability distribution, and forming a process-medium coupling map, pre-loading a target medium compensation parameter in an early warning state, calculating a time-varying interpolation weight based on a typical transition time length to carry out parameter dynamic fusion, and jointly applying the fused medium compensation parameter and a temperature drift compensation amount to flow measurement.

Description

Technical Field

[0001] This invention relates to the field of flow sensor data processing technology, and more specifically, to a flow sensor data processing method and system based on a multidimensional compensation model. Background Technology

[0002] In injection molding, MEMS thermal flow sensors are used to monitor the flow rate of mold coolant to achieve precise control of mold temperature. Mold temperature changes systematically according to the injection molding cycle (injection-holding pressure-cooling-mold opening). During this process, the cooling medium is switched at specific stages. For example, high-temperature heat transfer oil is used during the holding pressure stage to maintain the mold temperature, while cooling water is switched to accelerate cooling during the cooling stage. Therefore, the sensor must simultaneously address the periodic temperature drift caused by the process cycle and the changes in thermal response characteristics due to the switching of the cooling medium.

[0003] In the existing technology, the template compensation method for periodic temperature drift and the interpolation compensation method for medium transition state operate independently, without utilizing the temporal coupling law between the two in the process flow.

[0004] The drawback of existing technologies is that when the medium switching occurs at a specific phase of the process cycle, the two compensation mechanisms may produce superimposed errors or asynchronous responses. The temperature drift template queries the compensation amount according to a fixed phase, while the medium transition state compensation passively waits for changes in thermal response characteristics to trigger it. This independent operation mode leads to a temporary mismatch in compensation parameters at the moment of medium switching, resulting in a decrease in measurement accuracy during the transition period. Summary of the Invention

[0005] This invention provides a flow sensor data processing method and system based on a multidimensional compensation model, which solves the technical problem of decreased flow measurement accuracy caused by compensation parameter mismatch during media switching transition in related technologies.

[0006] This invention discloses a flow sensor data processing method based on a multidimensional compensation model, comprising: acquiring a historical sequence of ambient temperature signals; performing autocorrelation analysis on the historical sequence to determine the main process cycle; segmenting the historical sequence according to the main process cycle; calculating the temperature drift compensation amount for the differential temperature signals of each cycle segment and performing superposition averaging to generate a temperature drift cycle template; acquiring a historical sequence of thermal response characteristic signals; calculating the distance between adjacent characteristic points and performing moving average processing; determining a medium switching event when the moving average exceeds the transition state threshold; recording the switching time and medium type; calculating the phase of each medium switching event relative to the start of the process cycle; performing probability density estimation on the phase to generate a phase-switching probability distribution, forming a process-medium coupling spectrum; and calculating the current flow rate in real time. For the preceding process phase, the corresponding temperature drift compensation amount is queried from the temperature drift cycle template, and the phase-switching probability distribution is also queried. When the switching probability corresponding to the current phase exceeds the warning threshold, a warning state is entered. In the warning state, the target medium type and typical transition duration are pre-queried from the process-medium coupling spectrum, and the compensation parameter set of the target medium is pre-loaded. When the transition state start signal is detected, the time-varying interpolation weight is calculated based on the pre-queried typical transition duration, and the starting medium parameter and the target medium parameter are dynamically fused using the interpolation weight. The fused medium compensation parameter and the temperature drift compensation amount are jointly applied to the differential temperature signal, and the coupled compensated flow measurement value is output. The time-varying interpolation weight is the quotient of the time difference between the current time and the transition start time divided by the typical transition duration.

[0007] Furthermore, the step of performing autocorrelation analysis on the historical sequence to determine the main process cycle includes: calculating the cumulative sum of the environmental temperature signal multiplied by itself under different time delays to form an autocorrelation function; detecting the peak position of the autocorrelation function, and determining the time delay corresponding to the peak as the main process cycle.

[0008] Furthermore, the thermal response characteristic signal includes the temperature response time constant, steady-state temperature rise amplitude, and upstream and downstream temperature rise lag extracted from the differential temperature signal and the heating power signal.

[0009] Furthermore, the calculation of the phase of each medium switching event relative to the start of the process cycle includes: calculating the remainder after taking the modulus of the occurrence time of the medium switching event with respect to the main process cycle, and dividing the remainder by the main process cycle to obtain a normalized phase value.

[0010] Furthermore, the dynamic fusion of the initial medium parameters and the target medium parameters using the interpolation weights includes: using the interpolation weights as the weight coefficients of the target medium parameters, using the difference between the interpolation weights and 1 as the weight coefficients of the initial medium parameters, and performing a weighted summation of the two sets of parameters to obtain the fused medium compensation parameters.

[0011] Furthermore, the process-medium coupling map also includes the historical target medium type corresponding to each phase and the typical transition duration of historical statistics.

[0012] Furthermore, when the interpolation weight reaches 1 or the sliding mean of the thermal response characteristics is lower than the steady-state threshold, the transition state is determined to be over, the warning state is exited, and the actual transition duration of this transition is updated to the process-medium coupling map.

[0013] Furthermore, the probability density estimation employs a kernel density estimation method, which generates a continuous probability density curve by placing kernel functions at each phase sample point of the switching event and superimposing them.

[0014] Furthermore, the time-varying interpolation weights are nonlinearly mapped using an S-curve function, which makes the parameter changes smooth in the early and late stages of the transition, and accelerates the changes in the intermediate stage.

[0015] This invention discloses a flow sensor data processing system based on a multidimensional compensation model, used to execute the aforementioned method, comprising: a period analysis module for acquiring historical sequences of ambient temperature signals, performing autocorrelation analysis on the historical sequences to determine the main process cycle, and segmenting the historical data by cycle; a template generation module for calculating temperature drift compensation amounts for the differential temperature signals of each cycle segment and performing superposition averaging to generate temperature drift cycle templates; an event detection module for acquiring historical sequences of thermal response characteristic signals, detecting medium switching events through characteristic change rate analysis, and recording the switching time and medium type; and a coupling spectrum construction module for calculating the switching of each medium. The event phase relative to the start of the process cycle generates a phase-switching probability distribution, forming a process-medium coupling graph. The early warning judgment module calculates the current process phase in real time, queries the temperature drift compensation amount, and determines whether to enter a media switching early warning state. The parameter preloading module pre-queries the target medium parameters and typical transition duration from the process-medium coupling graph during the early warning state. The dynamic fusion module dynamically fuses the starting medium parameters and target medium parameters based on time-varying interpolation weights. The flow output module combines the fused medium compensation parameters with the temperature drift compensation amount and applies them to the differential temperature signal to output the coupled and compensated flow measurement value.

[0016] This invention extracts process cycle characteristics and establishes a temperature drift cycle template through autocorrelation analysis, establishes a medium switching event sequence through thermal response characteristic change rate detection, and then quantifies the temporal correlation between the two into a phase-switching probability distribution through kernel density estimation, forming a process-medium coupling map. Compared with independently operating temperature drift template compensation and medium transition state compensation, the process-medium coupling map enables the data processing system to proactively predict medium switching and preload target parameters at specific process phases, transforming transition state compensation from a passive response to active preparation. Because the occurrence phase of medium switching events in the process cycle has statistical regularity, a warning state can be entered before the actual switching occurs through the phase-switching probability distribution. Because the compensation parameter set and typical transition duration of the target medium have been pre-queried and loaded in the warning state, the calculation of time-varying interpolation weights can be performed based on historical statistical values ​​without waiting for the transition to end. Because the temperature drift compensation amount and the compensation parameter set take effect synchronously at the switching time, the factor of parameter mismatch during the transition period caused by asynchronous response when the two compensation mechanisms operate independently is overcome, thereby solving the technical problem of decreased flow measurement accuracy during the medium switching transition and achieving the technical effect of improving the measurement accuracy and response speed of flow sensors during medium switching. Attached Figure Description

[0017] Figure 1 This is a flowchart of the flow sensor data processing method based on a multidimensional compensation model according to the present invention. Detailed Implementation

[0018] In injection molding, MEMS thermal flow sensors are used to monitor the flow rate of mold coolant to achieve precise control of mold temperature. Mold temperature changes systematically according to the injection molding cycle (injection-holding pressure-cooling-mold opening). During this process, the cooling medium is switched at specific stages. For example, high-temperature heat transfer oil is used during the holding pressure stage to maintain the mold temperature, while cooling water is switched to accelerate cooling during the cooling stage. Therefore, the sensor must simultaneously address the periodic temperature drift caused by the process cycle and the changes in thermal response characteristics due to the switching of the cooling medium.

[0019] In existing technologies, template compensation methods for periodic temperature drift and interpolation compensation methods for media transition states operate independently, failing to utilize the temporal coupling between the two in the process flow. When media switching occurs at a specific phase of the process cycle, the two compensation mechanisms may produce superimposed errors or asynchronous responses: the temperature drift template queries the compensation amount according to a fixed phase, while the media transition state compensation passively waits for changes in thermal response characteristics to trigger it. This independent operation mode causes a temporary mismatch in compensation parameters at the moment of media switching, resulting in a decrease in measurement accuracy during the transition period.

[0020] The method of this embodiment includes the following steps; This embodiment provides a flow sensor data processing method based on a multidimensional compensation model, applied to an industrial process control system equipped with a MEMS thermal flow sensor. The system includes upstream and downstream temperature-sensitive elements for acquiring differential temperature signals, a temperature sensor for acquiring ambient temperature signals, a heating element and its power control module, and a data processing unit for executing this method.

[0021] Step 100: Obtain the historical sequence of ambient temperature signals, extract the main process cycle through autocorrelation analysis, and segment the historical data by cycle.

[0022] Acquire ambient temperature signal Autocorrelation analysis was performed on the long-term historical sequence:

[0023] in, It is the autocorrelation function. For time delay variables, The total number of samples in the historical sequence. For time indexing. Detect the peak position of the autocorrelation function to determine the main process cycle. At the same time, historical data is segmented according to the main process cycle to form a set of cycle segments. ,in This is the first periodic segment. This is the second periodic segment. This is the nth periodic segment.

[0024] Furthermore, in autocorrelation analysis, time delay variables The search scope is set to To ensure coverage of possible process cycle ranges, The total number of samples in the historical environmental temperature sequence is set, and the search range is set based on the Nyquist sampling theorem to avoid aliasing from affecting the accuracy of periodic detection.

[0025] Furthermore, the peak detection of the autocorrelation function is achieved through the following steps: first, exclude... The autocorrelation peak at that point, and then... Search within the range to satisfy and The local maxima point, the time delay corresponding to the first significant maxima point. As the main process cycle The significance criterion is that the autocorrelation coefficient at the maximum point is greater than 50% of the autocorrelation coefficient at time zero.

[0026] Furthermore, the periodic segmentation of historical data uses the local extreme points of the ambient temperature signal as the starting point of the period, and detects the ambient temperature sequence that satisfy... and The local minimum points are identified, and the data between two adjacent minimum points are considered as a periodic segment, thus forming a set of periodic segments.

[0027] In an application at a car dashboard injection molding production line, the system sampling frequency was 1Hz, collecting 28,800 sampling points of ambient temperature data over 8 consecutive hours. The production line process includes four stages: injection, holding pressure, cooling, and mold opening. The mold needs to maintain a high temperature during the holding pressure stage and needs to be cooled down rapidly during the cooling stage. The ambient temperature sensor is installed near the mold surface to reflect the periodic changes in the process.

[0028] Table 1. Historical ambient temperature data sample;

[0029] Autocorrelation analysis was performed on the historical series, with time delay variables... The search range is set to [0, 14400], and the autocorrelation function is calculated.

[0030] Table 2 Key results of autocorrelation analysis;

[0031] Autocorrelation function in The first significant maximum point appears at a certain point, with a relative amplitude of 50.0%, which meets the significance criterion (greater than 50% of the zero-delay autocorrelation coefficient). Therefore, the main process cycle is determined. Based on this period, the 28,800 sampling points were divided into 240 complete period segments, forming a set of period segments. Each period contains 120 sampling points.

[0032] Step 200: Calculate the temperature drift compensation amount for the differential temperature signal of each period segment and generate a temperature drift period template.

[0033] For each period segment Differential temperature signal within With ambient temperature The relationship is analyzed, and the temperature drift compensation amount of each segment is calculated. ,in The phase within the period.

[0034] Furthermore, temperature drift compensation amount The calculation method is as follows: in each period segment Within, for each phase point Differential temperature was established through linear regression. With ambient temperature Relationship ,in and For the regression coefficients, minimize the sum of squared residuals. We obtain, among which For sampling points within the periodic segment, Let the phase corresponding to the sampling point be denoted as , and then calculate the temperature drift compensation amount for that phase point as . ,in Indicates a periodic segment Mid-phase The corresponding ambient temperature measurement value, For reference ambient temperature, the average ambient temperature of each period is taken.

[0035] The compensation amounts for each period are superimposed and averaged to generate a temperature drift period template:

[0036] in, For normalized phase, The number of periodic segments, This refers to the period segment number. Temperature drift period template. It stores the typical temperature drift compensation values ​​for each phase during the process cycle.

[0037] Furthermore, normalized phase The calculation method is the ratio of the time within the cycle to the main cycle of the process, that is... This ensures that the phase value is within the range of [0,1), which facilitates the unified indexing of temperature drift cycle templates and phase-switching probability distributions.

[0038] Step 300: Obtain the historical sequence of thermal response characteristic signals, detect medium switching events through characteristic change rate analysis, and record the switching time and medium type.

[0039] Historical sequence of thermal response characteristic signals The thermal response characteristic signal reflects the current thermal properties of the fluid medium. Since the components of the characteristic vector have different dimensions, each component of the characteristic vector is normalized based on its range before calculating the Euclidean distance to eliminate the influence of dimensional differences on the distance calculation. The normalized Euclidean distance between adjacent characteristic points is then calculated:

[0040] Among them, subscript Indicates the index of the time sampling point. and They represent the first The and the first The thermal response feature vector at each sampling time.

[0041] Apply the moving average to the Euclidean distance sequence:

[0042] in, For the width of the sliding window, Indexed by the current time. For the first The characteristic rate of change at each time step. When the moving average... Exceeding the transition state threshold When this occurs, it is determined to be a media switching event, and the switching time is recorded. and the media type before and after the switch .

[0043] Furthermore, the medium type is identified by analyzing the current thermal response feature vector. The method involves nearest neighbor matching with a pre-established standard feature vector library for each medium type. Specifically, the Euclidean distance between the current feature vector and the feature vectors of each standard medium is calculated, and the standard medium type with the smallest distance is taken as the current medium type. The standard feature vector library is established by statistically averaging the thermal response features extracted from each known medium during steady-state operation.

[0044] Furthermore, the sliding window width The value is determined based on the system sampling frequency and the typical response time of medium switching. Specifically, it is the number of sampling points corresponding to the typical duration of thermal response characteristic changes during medium switching, so that the sliding window can cover the complete characteristic change process of medium switching.

[0045] Furthermore, the transition state threshold The value was determined by statistical analysis of the characteristic change rate during historical steady-state operation, specifically taken as three times the sum of the mean and standard deviation of the characteristic change rate during the steady-state period. ,in and These are the mean and standard deviation of the characteristic change rate during the steady state, respectively. The transition state threshold can effectively distinguish between steady-state fluctuations and characteristic changes caused by medium switching.

[0046] The production line uses heat transfer oil to maintain mold temperature during the injection-holding phase (phase 0.0-0.25), switches to cooling water near phase 0.25, and uses cooling water again during the cooling phase (phase 0.25-1.0) to accelerate cooling. The system extracts thermal response feature vectors. It contains three components: temperature response time constant. Steady-state temperature rise amplitude and upstream and downstream temperature rise lag After normalizing each component of the feature vector based on its range, the Euclidean distance between adjacent feature points is calculated, and a sliding window width is used. Perform moving average processing.

[0047] Table 3 Thermal response characteristic sequence and medium switching detection;

[0048] Based on the statistical analysis of the characteristic rate of change during steady state, the following calculations were performed: , Therefore, the transition state threshold At time index 390s, the moving average When the threshold is exceeded for the first time, the system determines it as a media switching event and records the switching time. Before the switch, the medium type was heat transfer oil (the Euclidean distance between the eigenvector and the standard heat transfer oil eigenvector was 0.12). After the switch, the medium type was cooling water (after 460s of steady-state recovery, the Euclidean distance between the eigenvector and the standard cooling water eigenvector was 0.09).

[0049] Step 400: Based on the media switching event sequence, calculate the phase of each switching event relative to the process cycle, generate the phase-switching probability distribution through kernel density estimation, and form a process-media coupling map.

[0050] Based on the detected media switching event sequence Calculate the phase of each switching event relative to the start of the process cycle, where This is the first media switching event. This is the second media switching event:

[0051] in, For the first The occurrence time of each handover event is determined. Kernel density estimation is performed on the phase of all handover events to generate a phase-handover probability distribution. The phase-switching probability distribution describes the probability of a medium switch occurring at each phase of the process cycle. Historical target medium type corresponding to each phase and typical transition duration Together they form a process-medium coupling pattern.

[0052] Furthermore, the typical transition duration corresponding to each phase. It is obtained by statistically analyzing the transition duration of media switching events that occurred near this phase in history. The specific method is as follows: for phase Filter out all occurrences in the phase interval Media switching events within, where The phase tolerance has a range of values. The specific value is determined based on the phase dispersion of historical switching events. When the phase distribution of switching events is concentrated, a smaller value is taken, and when it is dispersed, a larger value is taken. The time interval from the start of the transition state to the recovery of the steady state of these switching events is calculated, and the median of these time intervals is taken as the typical transition duration of the phase.

[0053] Furthermore, the historical target medium type corresponding to each phase. The target medium type was obtained by statistically analyzing the target medium type that occurred in the vicinity of this phase in the past, specifically the target medium type that appeared most frequently in this phase interval.

[0054] During 8 hours of continuous operation, the system detected a total of 240 media switching events. The normalized phase of each switching event relative to the start of the process cycle was calculated. .

[0055] Table 4. Statistical sample of media switching events;

[0056] Kernel density estimation was performed on the phase values ​​of 240 handover events using a Gaussian kernel function with a bandwidth parameter set to 0.03. The resulting phase-handover probability distribution was... In phase A significant peak is observed nearby, with a peak probability density of . For phases around 0.25 (phase tolerance) The switching events in the interval [0.20, 0.30] were statistically analyzed. There were 240 switching events in this interval, all targeting cooling water. The median transition time was 60 seconds. Therefore, the historical target medium type corresponding to this phase was determined to be $Type_{target}(0.25) = $"cooling water" with a typical transition time. .

[0057] Table 5 Key phase information of process-dielectric coupling spectrum;

[0058] The warning threshold is calculated as follows When the system's operating phase enters the [0.20, 0.30] interval, the probability density exceeds the warning threshold, and the system will enter the medium switching warning state.

[0059] Step 500: Obtain the current process phase in real time, query the corresponding temperature drift compensation amount from the temperature drift cycle template, and at the same time query the phase-switching probability distribution to determine the medium switching warning status.

[0060] Real-time calculation of the current process phase:

[0061] in, For the current moment, This indicates that the current position within the main process cycle is obtained by performing a modulo operation on the current moment.

[0062] From the temperature drift cycle template Query the temperature drift compensation amount corresponding to the current phase:

[0063] Furthermore, the temperature drift period template stores discrete compensation data points at fixed phase intervals, with the phase intervals being... Values ,in The number of discretization points for the template, typically ranging from [value range missing]. This ensures the template can fully capture the details of temperature drift changes throughout the process cycle. During the query, a linear interpolation method is used to obtain the compensation amount for any phase; specifically: find the current phase... The two closest template phase points and ,satisfy Calculate interpolation weights ,in , The compensation amount is obtained by indexing the phase point. .

[0064] Simultaneously query the phase-switching probability distribution ,when When the current state is determined to be a media switching warning state, among which... This is the warning threshold.

[0065] Furthermore, the warning threshold The value is determined based on the statistical characteristics of the phase-switching probability distribution, specifically set at 50% of the peak probability density of the phase distribution of historical media switching events. The warning threshold ensures that warnings are triggered in a timely manner in areas with a high probability of media switching, while avoiding too many false alarms in areas with a low probability.

[0066] Step 600: In the early warning state, pre-query the target medium type and typical transition time from the process-medium coupling diagram, and pre-load the compensation parameter set of the target medium.

[0067] When the system enters an early warning state, it pre-queries the target medium type corresponding to the current phase from the process-medium coupling map. and typical transition duration in historical statistics Preload the target medium's compensation parameter set The compensation parameter set for the target medium includes the corresponding thermal property parameters and flow calculation coefficients for the target medium.

[0068] Furthermore, the compensation parameter sets for each medium type are obtained through pre-calibration. The calibration process is as follows: under the condition of stable flow of a single medium, differential temperature signals, heating power, and the measured values ​​of a reference flowmeter are collected at different flow points. The specific heat capacity coefficient of the medium is obtained by fitting using the least squares method. thermal conductivity coefficient and flow-temperature response calibration coefficient The compensation parameter set for the medium is formed and stored in the parameter library.

[0069] Furthermore, the objective function for fitting the compensation parameter set is to minimize the mean square error between the reference flow rate value and the sensor-calculated flow rate value, i.e. ,in The number of flow points collected during calibration. For the first Reference flow meter readings at each flow point For parameter set The calculated flow rate value, The set of parameters to be optimized.

[0070] Step 700: Continuously monitor the rate of change of thermal response characteristics under the early warning state. When the transition state start signal is detected, record the transition start point and the initial medium parameters. Calculate the time-varying interpolation weight based on the pre-queried transition duration.

[0071] Continuously monitor the rate of change of thermal response characteristics under early warning conditions. When detected At that time, confirm the start of the transition state and record the transition start time. and initial medium parameters Typical transition duration obtained based on pre-query. Calculate the time-varying interpolation weights:

[0072] in, For the current moment, Let the elapsed time from the start of the transition state to the current moment satisfy the following condition: Weight The time-varying interpolation weights increase linearly with time and are used to control the fusion ratio between the initial and target medium parameters.

[0073] Furthermore, interpolation weights By constraining the minimum value function within the [0,1] interval, when the transition time exceeds the estimated typical transition time, the weight is automatically limited to 1, ensuring that the compensation parameters are completely switched to the target medium parameters and avoiding parameter fusion failure caused by the weight value exceeding the effective range.

[0074] In real-time monitoring during the 150th cycle of the process, the current moment... Corresponding process phase Querying the phase-switching probability distribution yields... The system has entered an early warning state. Pre-querying the process-medium coupling diagram reveals the target medium type to be cooling water, with a typical transition time of [missing information]. Compensation parameter set for preloaded cooling water The current set of compensation parameters for the heat transfer oil is as follows: .

[0075] exist At that time, the moving average was monitored. Confirm the start of the transition state and record the transition point. Based on a typical transition time of 60 seconds as determined by the pre-query, the system begins calculating time-varying interpolation weights and dynamically fusing parameters. The thermal conductivity coefficient in the parameter set... The synchronous linear transition from 0.145 W / (m·℃) to 0.628 W / (m·℃) is not listed in the table.

[0076] Table 6. Transitional dynamic compensation process;

[0077] During the transition state, the interpolation weights The specific heat capacity coefficient smoothly transitions from 2.35 kJ / (kg·℃) for heat transfer oil to 4.18 kJ / (kg·℃) for cooling water, increasing linearly from 0 to 1. The flow-temperature response calibration coefficient transitions from 0.0089 to 0.0152. Temperature drift compensation... Synchronously retrieved from the temperature drift cycle template, the temperature changes from 0.32℃ to 0.47℃ with the process phase. Differential temperature signal. The temperature rose from 12.48℃ to 24.63℃ during the medium switching process, reflecting the difference in thermophysical properties between cooling water and heat transfer oil. When the interpolation weight reaches 1.0, the compensation parameter is completely switched to the target medium parameter, the system exits the warning state and updates the process-medium coupling spectrum, adding the actual transition time of 60s to the historical statistics of phase 0.25.

[0078] Step 800: Dynamically fuse the compensation parameter set using interpolation weights, and apply the fused compensation parameter set together with the temperature drift period compensation amount to the differential temperature signal.

[0079] Using time-varying interpolation weights For the initial compensation parameter set With the target compensation parameter set Perform dynamic fusion:

[0080] Among them, the compensation parameter set , and Both are parameter vectors containing multiple parameter components. The above operation is a component-wise linear interpolation operation, that is, interpolation calculation is performed on each parameter component in the parameter set separately.

[0081] The fused compensation parameter set Compensation amount for temperature drift cycle Combined application to differential temperature signals , to perform flow calculation.

[0082] Furthermore, the compensation parameter set Specific heat capacity coefficient of the medium thermal conductivity coefficient and flow-temperature response calibration coefficient During dynamic fusion, linear interpolation is performed on each parameter component separately, i.e. The fusion method for other parameter components is the same.

[0083] Step 900: Output the flow measurement value after process cycle and medium state coupling compensation. When the transition state ends, exit the early warning state and update the statistics of the process-medium coupling graph.

[0084] Output the flow measurement value after coupling compensation of process cycle and medium condition:

[0085] in, For heating power, The flow calculation function is based on the thermal balance principle of a thermal flow sensor, and uses the compensated differential temperature. Heating power and compensation parameter set Mapped to flow measurement values. When the transition state ends (satisfying...) or ,in When the threshold value is reached (the steady-state threshold), the system exits the warning state and updates the actual transition duration and other information to the statistics of the process-medium coupling graph.

[0086] Furthermore, the update method for the process-medium coupling map is as follows: record the phase of the occurrence of this media switching event. and actual transition duration ,in The end time of the transition state, At the start of the transition state, this data point is added to the phase. The typical transition duration for that phase is then recalculated from the corresponding set of historical switching events. A sliding window approach is used to retain data from the most recent N_window switching events to achieve adaptive updates of the map. The range of values ​​is The specific value is determined based on the stability of the process. When the process changes frequently, a smaller value is taken to improve the adaptive speed, and when the process is stable, a larger value is taken to improve the statistical reliability.

[0087] Furthermore, the flow calculation function The specific form is: ,in This is the calibration factor for the current medium's flow-temperature response. Given the specific heat capacity coefficient of the current medium, the flow rate calculation function is derived based on the relationship between heating power and fluid mass flow rate, and the product of specific heat capacity and temperature rise in the heat balance equation. When applying the flow rate calculation function, the following conditions must be met: The conditions are met to ensure the validity of the square root operation.

[0088] Furthermore, steady-state threshold The value is lower than the transition state threshold. Specifically, the value is twice the sum of the mean and standard deviation of the characteristic rate of change during the steady-state period, i.e. The steady-state threshold is used to determine that the medium switching transition process has been completed and a new steady state has been entered. By setting a steady-state threshold lower than the transition state threshold, a hysteresis characteristic is formed to avoid frequent jitter in the transition state determination.

[0089] It should be noted that the thermal response characteristic signal in step 300 above... The thermal response feature vector refers to the feature vector extracted from the differential temperature signal and the heating power signal that can characterize the current thermal properties of the fluid medium. The thermal response feature vector can include components such as the time constant of the temperature response, the steady-state temperature rise amplitude, and the upstream and downstream temperature rise lag.

[0090] Furthermore, the method for extracting each component of the thermal response feature vector is as follows: the time constant of the temperature response is obtained by exponentially fitting the differential temperature response curve after a step change in heating power. Obtain the time constant ,in is the base of the natural logarithm. This represents the elapsed time since the step change in heating power. This represents the steady-state temperature rise amplitude. The time constant parameter to be fitted is used, and the fitting is achieved using the least squares method. The steady-state temperature rise amplitude is obtained by calculating the average of the differential temperatures within a certain time window after the heating power stabilizes. The time window length for calculating the steady-state temperature rise amplitude is set to [value missing]. to ( (The time constant is used to ensure that the temperature response has reached more than 95% of the steady-state value; the upstream and downstream temperature rise lag is obtained by calculating the time difference between the peak times of the temperature response of the upstream and downstream temperature-sensitive elements.)

[0091] It should be noted that the kernel density estimation in step 400 above is a non-parametric probability density estimation method. It generates a smooth probability density curve by placing a kernel function at each sample point and superimposing them. Compared with the histogram method, it can more accurately reflect the continuous characteristics of the phase distribution.

[0092] The input to the aforementioned kernel density estimation is the set of normalized phase values ​​of all media switching events relative to the start of the process cycle. ,in The total number of media switching events; the output is defined within the phase interval. Continuous probability density function on The probability density function describes the probability density of medium switching occurring at each phase of the process cycle.

[0093] The input for the aforementioned autocorrelation analysis is a time series of ambient temperature signals. ,in The ambient temperature at time 1. The ambient temperature at the second moment. Let K be the ambient temperature at time k. The total number of sampling points in the time series is given, and the output is the main process cycle. and the set of periodic segments after periodic segmentation .

[0094] In this embodiment of the application, based on step 600, the following optimization process may also be included: when the duration of the warning state exceeds the preset duration but no transition state start signal is detected, the warning confidence is gradually reduced and the cache occupation of preloaded parameters is reduced to avoid resource waste caused by false warnings.

[0095] In this embodiment of the application, based on step 800, in order to further improve the compensation accuracy during the transition period, time-varying interpolation weights are used. The calculation can use nonlinear functions instead of linear calculations. For example, an S-curve (Sigmoid function) can be used to smooth the transition, making the parameter changes more gradual in the early and late stages of the transition, and faster in the middle stage, which is more in line with the physical characteristics of the actual medium replacement process.

[0096] The flow sensor data processing method based on the multidimensional compensation model provided in this embodiment extracts process cycle characteristics and establishes a temperature drift cycle template through autocorrelation analysis, establishes a medium switching event sequence through thermal response characteristic change rate detection, and then quantifies the temporal correlation between the two into a phase-switching probability distribution through kernel density estimation, forming a process-medium coupling spectrum.

[0097] Compared to independently operating temperature drift template compensation and media transition state compensation, the process-media coupling graph enables the data processing system to proactively predict media switching and preload target parameters at specific process phases, transforming transition state compensation from a passive response to active preparation. Because the occurrence phases of media switching events within the process cycle exhibit statistical regularity, a warning state can be entered before the actual switching occurs through the phase-switching probability distribution. Since the target medium's compensation parameter set and typical transition duration have been pre-queried and loaded in the warning state, the calculation of time-varying interpolation weights can be performed based on historical statistical values ​​without waiting for the transition to end. Because the temperature drift compensation amount and the compensation parameter set take effect synchronously at the switching moment, the factor of parameter mismatch during the transition period caused by asynchronous responses when the two compensation mechanisms operate independently is overcome, thus solving the technical problem of decreased flow measurement accuracy during media switching transitions.

Claims

1. A method for processing flow sensor data based on a multidimensional compensation model, characterized in that, Includes the following steps: The historical sequence of ambient temperature signal is acquired, and autocorrelation analysis is performed on the historical sequence to determine the main process cycle. The historical sequence is segmented according to the main process cycle, and the temperature drift compensation amount is calculated for the differential temperature signal of each cycle segment and then superimposed and averaged to generate a temperature drift cycle template. Acquire the historical sequence of thermal response characteristic signals, calculate the distance between adjacent characteristic points and perform moving average processing. When the moving average exceeds the transition state threshold, it is determined to be a medium switching event. Record the switching time and medium type. Calculate the phase of each medium switching event relative to the start of the process cycle, perform probability density estimation on the phase to generate a phase-switching probability distribution, and form a process-medium coupling map; The current process phase is calculated in real time. The corresponding temperature drift compensation amount is queried from the temperature drift cycle template. At the same time, the phase-switching probability distribution is queried. When the switching probability corresponding to the current phase exceeds the warning threshold, the warning state is entered. In the early warning state, the target medium type and typical transition time are pre-queried from the process-medium coupling map, and the compensation parameter set of the target medium is pre-loaded; When a transition state initiation signal is detected, a time-varying interpolation weight is calculated based on the typical transition duration pre-queried, and the initiation medium parameters and target medium parameters are dynamically fused using the interpolation weight; The fused medium compensation parameters and temperature drift compensation are combined and applied to the differential temperature signal to output the coupled and compensated flow measurement value. The time-varying interpolation weight is the quotient of the time difference between the current time and the transition start time divided by the typical transition duration.

2. The method according to claim 1, characterized in that, The autocorrelation analysis of the historical sequence to determine the main cycle of the process includes: The autocorrelation function is formed by multiplying the ambient temperature signal with itself at different time delays. The peak position of the autocorrelation function is detected, and the time delay corresponding to the peak is determined as the main process cycle.

3. The method according to claim 1, characterized in that, The thermal response characteristic signal includes the temperature response time constant, steady-state temperature rise amplitude, and upstream and downstream temperature rise lag extracted from the differential temperature signal and heating power signal.

4. The method according to claim 1, characterized in that, The calculation of the phase of each media switching event relative to the start of the process cycle includes: The remainder after taking the modulus of the moment of the media switching event with respect to the main process cycle is calculated, and the remainder is divided by the main process cycle to obtain the normalized phase value.

5. The method according to claim 1, characterized in that, The dynamic fusion of the initial medium parameters and the target medium parameters using the interpolation weights includes: Using the interpolation weight as the weight coefficient of the target medium parameter, and the difference between the interpolation weight and 1 as the weight coefficient of the starting medium parameter, the two sets of parameters are weighted and summed to obtain the fused medium compensation parameters.

6. The method according to claim 1, characterized in that, The process-medium coupling map also includes the historical target medium type corresponding to each phase and the typical transition time of historical statistics.

7. The method according to claim 1, characterized in that, When the interpolation weight reaches 1 or the sliding mean of the thermal response characteristics is lower than the steady-state threshold, the transition state is determined to be over, the warning state is exited, and the actual transition duration of this transition is updated to the process-medium coupling map.

8. The method according to claim 1, characterized in that, The probability density estimation employs a kernel density estimation method, which generates a continuous probability density curve by placing kernel functions at each phase sample point of the switching event and superimposing them.

9. The method according to claim 1, characterized in that, The time-varying interpolation weights are nonlinearly mapped using an S-curve function, which makes the parameter changes smooth in the early and late stages of the transition, and accelerates the changes in the intermediate stage.

10. A flow sensor data processing system based on a multidimensional compensation model, used to execute the method according to any one of claims 1-9, characterized in that, include: The cycle analysis module is used to acquire historical sequences of ambient temperature signals, perform autocorrelation analysis on the historical sequences to determine the main process cycle, and segment the historical data by cycle. The template generation module is used to calculate the temperature drift compensation amount of the differential temperature signal in each period segment and perform superposition and averaging to generate a temperature drift period template. The event detection module is used to acquire historical sequences of thermal response characteristic signals, detect media switching events through characteristic change rate analysis, and record the switching time and media type. The coupling graph construction module is used to calculate the phase of each medium switching event relative to the start of the process cycle, generate the phase-switching probability distribution, and form a process-medium coupling graph. The early warning judgment module is used to calculate the current process phase in real time, query the temperature drift compensation amount, and determine whether to enter the medium switching early warning state. The parameter preloading module is used to pre-query the target medium parameters and typical transition times from the process-medium coupling diagram in the early warning state. The dynamic fusion module is used to dynamically fuse the starting medium parameters and the target medium parameters based on time-varying interpolation weights. The flow output module is used to combine the fused medium compensation parameters with the temperature drift compensation amount and apply them to the differential temperature signal to output the coupled and compensated flow measurement value.

Images