Flowmeter-oriented embedded multi-parameter fusion calibration control method and system

Abstract

The invention discloses a flowmeter-oriented embedded multi-parameter fusion calibration control method and system, relates to the technical field of industrial measurement and embedded control, and is used for solving the problem of unstable calibration caused by misalignment of multi-link flow signals, compensation lag and fusion oscillation. The method relates to synchronous acquisition, pretreatment and calibration of differential pressure type, ultrasonic wave and thermal type mass measurement links and environment parameters such as temperature, pressure and density. All-channel unified time scales are established through ternary binding, filtering and denoising, zero snapshot and alignment rules, a stability scoring mechanism is constructed in combination with offset behavior recognition, event classification and feature aggregation, and dynamic switching of a compensation path and a fusion strategy is driven. In a double-path fusion structure, table look-up compensation, model estimation, multi-algorithm parallelism and fusion track recording are adopted, version traceability and calibration result closed-loop control are achieved, and the stability and rechecking performance of mass flow output under complex disturbance are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of industrial measurement and embedded control technology, and more specifically, to an embedded multi-parameter fusion calibration control method and system for flow meters. Background Technology

[0002] Existing flowmeter systems generally face problems such as poor synchronization of multiple physical channels, inconsistent measurement link characteristics, frequent environmental disturbances, and poor output stability. Especially in industrial settings, differential pressure, ultrasonic, and thermal flowmeters have different response rates and sensitivities. When the measured medium is subject to perturbations such as heat exchange, backflow, or short-lived bubbles, it often leads to problems such as signal window errors, compensation lag, or false triggering of fusion strategies. In addition, environmental parameters such as temperature, pressure, and density themselves have hysteretic or nonlinear response characteristics. Inconsistent sampling cycles between these parameters and primary signals, or mismatches in compensation models, can easily lead to systematic errors in mass flow rate calculations.

[0003] Traditional methods often employ fixed compensation coefficients and fusion weights, lacking in-depth modeling of the temporal characteristics of measurement events. This makes it difficult to effectively distinguish disturbance sources, identify dominant channels, or dynamically adjust fusion strategies, easily leading to unstable outputs during rapid temperature and pressure changes or measurement link drift. Furthermore, existing systems often lack state recording and version encapsulation for the entire measurement and compensation path, making it difficult to determine the source of anomalies and version change paths in post-event tracing, and also hindering adaptive fusion optimization.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an embedded multi-parameter fusion calibration control method and system for flow meters, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a preferred embodiment, it includes: Activate the flow measurement signal and environmental parameter channels in sequence, set the communication and pin positions and write the three-element binding, and sample differential pressure, forward and reverse flow first arrival, bridge differential judgment bridge arm balance, and collect temperature and pressure to calculate density according to two diameters; Extract frame information of clean sequences for each channel, monitor and mark alignment breaks, compensate for feedback re-entry, and sensitivity hysteresis, and generate a set of rules; after the classification labels and source paths are clear, group them according to channel type, event label and time order, calculate stability scores and select the dominant path; Differentiated convergence is applied to the preprocessing paths of different measurement links. If the zero drift exceeds the threshold or the bridge arm is unbalanced, the calculation state is restricted and the write channel is frozen. After the anomaly is under control, the compensation path and the multi-algorithm path are opened in parallel.

[0008] In a preferred embodiment, the flow measurement signal channel and the environmental parameter channel are activated sequentially; communication parameters and interface pins are set according to the data channel registration table, the timestamp, device identifier and channel number are written in a three-element binding, the temperature compensation coefficient, density correction relationship and range segmentation correction curve are loaded, and port level and pull-up or pull-down checks, communication handshake detection and register loopback verification are performed.

[0009] In a preferred embodiment, the differential pressure digital sample, the first arrival time of the forward and reverse currents, and the bridge differential signal are sampled to determine the balance of the bridge arms; Temperature and pressure are collected by a thermistor and a pressure transmitter, and density is calculated according to the state equation or calibration curve. Then, the system polls at a uniform sampling period and performs noise reduction and blanking in a ring buffer.

[0010] In a preferred embodiment, after the preprocessing results are produced, the source identifier, timestamp, event pin index, filter version and sampling phase, lookup table or model interval and version of each clean sequence are extracted and written into the black box; then, three types of offsets are monitored and marked: alignment break, compensation feedback reentry, and sensitivity hysteresis; when the offsets coincide or persist, a set of rules is generated accordingly.

[0011] In a preferred embodiment, the rule set is as follows: Relative timescales are reconstructed using ultrasound first arrival or differential pressure second-order zero crossover as needle points and delayed compensation injection is performed. Density back-writing is delayed once in the critical region and temporary density proxy is generated by using arrival time difference / envelope half-width difference / single point temperature as observation quantity. After backfilling the record, thermal path weight is adjusted in the fusion unit and process noise is tightened. After the criteria are met, segmented recovery is performed, and parameter source and version are recorded simultaneously. Then, according to the rule set, the classification and labeling are performed as acoustic first type, cross-regional jitter type, viscosity hysteresis type, and feedback re-entry type.

[0012] In a preferred embodiment, after the classification labels and source paths are clear, the event sequence groups are divided according to channel type, event label and time order; the mean error window, mean consistency kernel, interval switching frequency, freeze trigger rate and fusion branch switching rate are extracted for each group and feature vectors are generated by sliding aggregation; then the stability score is calculated and sorted according to the feature vectors, the high-scoring group is selected as the dominant path and the target alignment reference, and the low-scoring and low-conflict groups are subjected to flexible registration and small-step weight climbing. Simultaneously, the synchronous reference conflict mapping cluster implements dynamic calibration and strategy adjustment. When a high-conflict trajectory is hit, it performs strategy reverse compression, suspends regular weighted fusion, adopts a conservative output of dominant path and historical smoothing, and implements label dilution and time gap protection for conflict paths.

[0013] In a preferred embodiment, envelope extraction, peak latching, and window smoothing are performed on the ultrasound channel; zero-point residuals are calculated by sliding window regression in the static pressure segment for the differential pressure channel; the bridge arm ratio and reference difference are statistically tested and time-stamped for the thermal bridge channel; zero-point and equilibrium state labels are generated for event classification and stability scoring; and when the preprocessing determines that the zero-point drift exceeds the threshold or the bridge arm is unbalanced, a restricted calculation state is set, the write channel is frozen, and only the basic output is retained.

[0014] In a preferred embodiment, two fusion calibration paths are activated in parallel under abnormally controlled conditions: The compensation path is based on table lookup or model injection of temperature, pressure, and density compensation, and the version and triggering reason are recorded.

[0015] The multi-algorithm path runs dual solvers in each channel, compares with historical output and operating condition thresholds, performs fusion or masking, and records branches, weights and masking reasons, and writes them into the fusion trajectory to support sequence grouping, stability scoring and conflict mapping.

[0016] In a preferred embodiment, it includes: an acquisition and density calculation module, an event classification and scoring module, a convergence and strategy execution module, and a data storage module, with signal connections between the modules; The data storage module is used to store all data during the storage process; The acquisition and density calculation module is used to sequentially activate the primary flow measurement signal and environmental parameter channels, set the communication and pin positions and write the three-element binding, and sample differential pressure, forward and reverse flow first arrival, bridge differential judgment bridge arm balance, and acquire temperature and pressure to calculate density according to two diameters; The event classification and scoring module is used to extract frame information of the clean sequence of each channel, monitor and mark alignment breaks, compensate for feedback re-entry, sensitivity hysteresis and generate rule sets; after the classification labels and source paths are clear, the channels are grouped by channel type, event labels and time order and stability scores are calculated to select the dominant path; The convergence and strategy execution module is used to perform differentiated convergence on different measurement link preprocessing paths, determine zero drift exceeding the threshold or bridge arm imbalance to set the computation state and freeze the write channel, and after the anomaly is under control, the compensation path and multi-algorithm path are opened in parallel.

[0017] The technical effects and advantages of the embedded multi-parameter fusion calibration control method and system for flow meters of this invention are as follows: This invention achieves unified acquisition and preprocessing of three types of measurement links and three types of environmental parameters across all channels. Combined with frame-level alignment, zero-point snapshots, and version tracking mechanisms, it ensures consistency of all input data across time scales and compensation paths. By identifying abnormal events such as alignment breaks, compensation feedback re-entry, and sensitivity lag through behavioral markers, it establishes an event classification and grouping mechanism. For each group, it calculates indicators such as error window degree and consistency kernel to construct a stability score, dynamically selecting the dominant path and fusion strategy during the fusion phase.

[0018] At the fusion execution level, a dual-path strategy is introduced. On the one hand, physical quantities are corrected through temperature / pressure / density lookup table compensation and model estimation. On the other hand, the credibility of the fusion results is improved through parallel dual solvers, fusion trajectory recording and Kalman fusion optimization. When the compensation or fusion path deviates, shielding, freezing and strategy switching are implemented.

[0019] By recording the version and triggering details of all compensation / fusion paths through a black box mechanism, the system supports runtime backtracking and version re-inspection, further improving the stability, transparency, and engineering feasibility of the solution. It is particularly suitable for online quality flow control applications with strong dynamic disturbances and significant link asynchrony in industrial scenarios. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the embedded multi-parameter fusion calibration control method and system module for flow meters according to the present invention.

[0021] Figure 2 This is a timing diagram of the embedded multi-parameter fusion calibration control method and system for flow meters according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example: This invention discloses an embedded multi-parameter fusion calibration control method for flow meters, such as... Figure 2 As shown, it includes: First, the acquisition activation process of the flow measurement signal channel and the environmental parameter channel is triggered sequentially through the multi-source data channel scheduling mechanism. For each data channel, the parameters of electrical communication protocols such as Modbus-RTU / HART / RS-485 are written to the disk and the interface pin mapping is set according to the predetermined number, communication protocol and interface type in the data channel registration table. Before the data is introduced, the acquisition timestamp is written, and the three-element binding of device identifier and channel number is completed. At the same time, the temperature compensation coefficient table, density correction relationship and range segment correction curve solidified by calibration test are loaded.

[0024] Subsequently, through physical port wiring including port level and pull-up or pull-down checks, communication handshake response detection and register address read / write loopback verification operations, the initial operating state of the channel logic being online, the hardware connectivity being correct, and the parameters being readable is obtained, that is, the channel is online and the parameters are valid.

[0025] In terms of signal acquisition, the activation of the full signal is achieved by configuring differentiated sampling paths for the three types of measurement links: differential pressure, ultrasonic, and thermal mass.

[0026] Specifically, firstly, the differential pressure flow meter uses the analog differential pressure signal output by the differential pressure transmitter and the digital sample obtained by the A / D conversion module. By loading the sampling range with the upper and lower limits of the range and the reference voltage coefficient, and binding the current clock source to perform zero-point snapshot, the raw differential pressure data sequence with absolute time scale is obtained. Secondly, the ultrasonic flow meter acquires time history data through the bidirectional propagation link of the forward / backward transducer, and obtains the paired forward / backward arrival times through TTL threshold triggering, counter clearing and first pulse capture operations; Third, the thermal mass flow meter obtains the differential signal of the bridge through the differential measurement path composed of the heating bridge and the temperature sensing bridge. After the signal is amplified, it is sampled by the A / D interface, and the differential voltage reading and bridge balance status mark information are obtained by comparing the bridge arm balance potential difference with the reference channel ratio.

[0027] Regarding the acquisition of environmental parameters, temperature signals are obtained through NTC thermistors that have completed traceability calibration and the real-time temperature value is calculated according to the thermistor characteristic curve; pressure and static pressure signals are obtained through diaphragm pressure transmitters to acquire analog outputs and are sampled and calibrated, and the corresponding readings are calculated according to the voltage-pressure mapping relationship defined by the piezoresistive strain coefficient; density parameters are not directly sampled, but are calculated by inputting the above real-time temperature and pressure into the medium state equation model or by obtaining the back-calculation result through the calibrated density curve.

[0028] The solution for ρt follows one of two methods, and the method and version used are written into the black box. The scope of the state equation, ρt= (Pt,Tt)), which records the EOS type, parameter source, and version; Calibration curve aperture, ρt= (Pt,Tt|interval number,version), use linear or piecewise spline interpolation, extrapolation is prohibited; record the interpolation method, interval number and version.

[0029] Among them, to avoid To become a black box, first register it in the data channel registration form. The system explicitly provides four aperture types: {Ideal, SRK, PR, Custom}, with PR as the default. When the compressibility factor obtained from the static section calculation satisfies |Z−1|≤0.02, the system automatically switches to the Ideal aperture. SRK is prioritized when the medium description indicates a light hydrocarbon gas phase. If an approved medium-specific residual Helmholtz free energy model is provided, it is designated as Custom (model version and parameter source are attached). The appropriate model is included in each density solution. The parameter group and valid fields are recorded in the black box. If (P,T) exceeds the selected EOS valid fields, extrapolation is prohibited and the data is rolled back to table-driven interpolation, with the reason for rollback and timestamp specified.

[0030] Next, the table-driven interpolation uses bidirectional separation and constraint cubic spline aperture within the cell: the nodes of the T-axis and P-axis are generated according to regular mesh or iso-density sensitive adaptive mesh with ΔT=5~10K and ΔP=0.05~0.20MPa. For a single axis, C² consecutive cubic splines are first generated and Hyman monotonicity constraints are applied to prevent overshoot. If the local monotonicity is still violated, it is automatically downgraded to PCHIP. The boundary conditions default to natural boundaries. When the first derivative of the endpoint is measurable, clamped boundaries are used. The two-dimensional values ​​are interpolated one-to-one, and leave-one cross-validation is performed for each mesh cell to ensure that the relative MAE ≤ 1.0%. If the standard is not met, the nodes are refined or the calculation is rolled back to EOS. All node numbers, boundary types, error statistics and version numbers are written into the black box.

[0031] Among them, the calibration data of the density correction curve is divided into temperature / pressure working areas according to the medium: the T interval should cover at least three segments and the endpoints should fall within 85% of the equipment's achievable limit; the P interval should have at least two levels and include the static pressure baseline segment; the total sample size of a single medium should be ≥500 sets and each sub-interval should have ≥100 sets; and the data source should be limited to two channels: water tank / constant temperature and pressure vessel and on-site reinjection.

[0032] The black box is defined as a set of operational traceability records. The minimum set includes {timestamp, device identifier, channel number, version, interval number, caliber, trigger reason, and status flag}. It is written in the order of measurement cycles and does not overwrite. It is used for playback and version review.

[0033] When (P,T) hits an interval endpoint, it is processed according to the endpoint value; when the interval number or version is inconsistent, it reverts to the last valid version and writes it into the black box; if it is still unusable, it switches to the state equation approach and records the reason.

[0034] The three types of environmental parameters are then polled using the same sampling period as the primary signal. First-order denoising and transition blanking operations are performed on the continuous samples using an anti-jitter ring buffer to obtain temperature, pressure, and density readings that are aligned with the primary signal frame level on the acquisition time axis.

[0035] Subsequently, at the acquisition end, a preprocessing stage is entered, where a global time stamp is established using a unified hardware timer to align the data from each channel on the same time grid. When the alignment residual continuously exceeds the threshold, a resynchronization is performed, and the resynchronization time and reason are written into the black box.

[0036] Based on this, by loading each channel with the registered and confirmed FIR / IIR filter type and order, pass / stopband boundary and coefficient set, and performing DC component subtraction operation with zero-point snapshot as reference, a set of multi-channel clean sequences that are denoised, debiased and time-scaled are obtained. The parameters and versions used in this preprocessing are written into the black box, and the zero-point snapshot time is recorded.

[0037] Meanwhile, to meet feasibility requirements, the implementation details for each channel are recorded, including sampling rate fs, FIR / IIR and order, pass / stop band boundaries, and coefficient version; in fixed-point implementation, the coefficient quantization bit width and data path bit width are written.

[0038] It should be noted that in field conditions involving heat exchange, bypass mixing, and reflux, there is often an asynchronous disturbance chain upstream of the measured section, characterized by short-life phase transitions of microbubbles, temperature step transitions, and slight valve position swings. In this situation, the ultrasonic link will initially respond rapidly to changes in acoustic compressibility via ToF transitions and envelope patterns. The differential pressure link, influenced by throttling devices and stagnant water sections, exhibits a delayed, narrow-band breathing wave. The temperature / static pressure channel, affected by heat and pressure transfer paths, will experience further hysteresis. If frame-level alignment is achieved using only a unified hardware timescale during preprocessing, and the clean sequences of each channel are directly pushed into the compensation and fusion paths, event slices from different physical moments will be mixed into the nominal simultaneous frame. The compensation amount for density lookup table transitions may align to the previous ultrasonic transient, and the sensitivity correction of the thermal bridge arm may fall on the boundary of the differential pressure misalignment. This results in a combined effect within the same computational grid: misaligned injection of compensation and compensated windows, imbalance of fusion branch criteria, and false triggering of hysteresis / freeze thresholds, affecting the stable output of instantaneous mass flow.

[0039] In existing solutions, frame-level alignment is assumed to be physical event synchronization. Explicit semantic paths are not established for the source, arrival order, and interval switching of events across channels. Furthermore, intermediate quantities such as misaligned window metrics, consistency kernels, and event health indicators are not introduced. This leads to situations where, in typical windows where microbubble acoustic transients arrive first, density lags across critical zones, and thermal sensitivity lags again, compensation values ​​are misaligned, fusion misjudgments result in reduced weighting, and excessively dense short-term freezes occur. Moreover, because the output messages themselves lack encapsulated identifiers for event source trajectories, policy triggers, and release conditions, version review and black-box verification stages struggle to distinguish whether the issue stems from asynchronous operating conditions or path algorithm selection, leading to biased judgments in maintenance and cause analysis.

[0040] Therefore, in this embodiment, after the preprocessing results are produced and before compensation and fusion, firstly, after the clean sequence of each channel is generated, the source and temporal metadata are extracted and retained synchronously, including the channel source identifier, frame timestamp, first arrival / zero crossover / imbalance threshold event needle index, filter version and sampling phase, SNR and envelope morphology parameters, near window gradient, lookup table / model interval number and its version, and written into the frame metadata and black box; Next, event behavior tagging is completed before structured fields are generated. Within a single computation cycle, the following three types of post-sample offset phenomena, which are only engineering-significant within the current computational chain, are continuously captured; The first type is alignment break, which means that the ultrasound ToF / envelope transitions first, but the differential pressure and temperature / pressure compensation do not produce a corresponding response in the adjacent segments. The identification feature is that the event needle time difference exceeds the threshold and the cross-channel correlation decreases. The second type is compensation feedback reentry, which means that the density repeatedly crosses segments in the critical interval, and the hysteresis freeze and version write-back overlap within the short window. This is manifested as the frequency of table lookup interval switching and the freeze trigger count increasing synchronously. The third type is sensitivity hysteresis, which means that the response of the thermal bridge to changes in pressure / viscosity lags behind that of differential pressure / environmental parameters. The net imbalance of the bridge arms only converges at the end of the event, forming a hysteresis band.

[0041] When the aforementioned offsets coincide or persist within a single event window, the system enters the rule generation channel. This channel does not rely solely on a fixed compensation sequence but is driven by the tensor mapping of the alignment break zone, feedback reentry zone, and hysteresis band to dynamically generate a set of rules, including but not limited to: Alignment is triggered by a breakage event. The first arrival event of the ultrasonic channel (US first arrival) or the second-order zero crossover of the differential pressure Δp of the differential pressure channel is used as the needle point. The relative time scale within the event is reconstructed, and the compensation injection is delayed until alignment is completed. The relative timescale remapping caliber is as follows: Set reference needle point The time remapping of channel c is The calculation of the compensation amount and the amount being compensated is in... Completed within the coordinate system; only if | - |≤ And ε≤ And κ≥ Alignment is considered complete. The source and version are written into the black box.

[0042] Compensation feedback reentry triggers temporary density proxy, delaying one density write-back in the critical region, and using a low-dimensional observer based on Time-of-Flight (ToF), envelope left and right half-width difference, and single-point temperature for estimation. and Participate in one compensation, and after the correction is lifted, backfill with the actual density to rectify the deviation; Temporary density proxies can compute expressions. Let the difference in the left and right half-widths of the envelope be ΔHW, the temperature at a single point be T, and the Time-of-Flight (ToF) difference be δτ, specifically based on the formula. ; = or = ; in Derived from reinjection experiments or type tests and written into the version; this version only uses... After participating in a compensation process, the difference and its cause are written into the black box after the actual density is written back.

[0043] Subsequently, without altering the computational framework of the temporary density proxy, a training / reservation stratified set was first constructed by merging two types of samples: the flume-type test and the field recharge. Then, least squares estimation was used. and The initial value is obtained, and when the residual has a long tail, Huber robust regression is switched once to achieve quadratic convergence. At the same time, R², MAE, RMSE and 95% confidence interval are written into the black box through 1,000 bootstraps; among them The physical dimension is K / normalized half-width. The physical dimension is K / μs. Under normal working conditions, the qualified interval obtained by the above fitting is limited to k_T∈[0.40,0.70] and b_T∈[0.02,0.06]. When the margin set MAE exceeds 1.10 times the training set MAE or R² is less than 0.95, the current parameter is determined to be invalid and the previous valid version is rolled back. At the same time, the reason for rollback, sample size and timestamp are recorded. To ensure reproducibility, the training / holdout ratio is fixed at 8 / 2. The random seed and the fitted script version are written into a black box, and parameter stability is checked every 24 hours during a quiet period. If Δρ = If the threshold is continuously exceeded within the sliding window, the refit evaluation queue is triggered and the compensation update is frozen for 1 second.

[0044] Specifically, the original ultrasound signals from both upstream and downstream currents are debanded using a fourth-order zero-phase IIR bandpass at ±20% of the carrier frequency, followed by amplitude normalization and linear detrending to eliminate slow drift; based on this, the first arrival time t0 is located using STA / LTA and the first arrival window is set accordingly. ],For example, Take the larger of the eight carrier cycle duration and 0.8 ms, and use the Hilbert envelope e(t) = |Hilbert{ | Obtain peak value within the first window with half-high threshold Therefore, the left / right threshold crossing times are calculated. And give the half-width HW= Then, using the historical steady-state median Normalization yields And only after applying five-point median smoothing to ΔHW did it enter. ; At the same time, for downstream The normalized cross-correlation is calculated on the relevant window W centered at the first peak, and quadratic parabolic subsampling interpolation is performed in the neighborhood of the main peak to obtain the result. Then use exponential moving average Obtain δτ, where α∈[0.6,0.8], default 0.7; Next, zero drift is subtracted using the median of the static segment defined by the Allan minimum variance window to eliminate slow-varying bias; when the cross-correlation primary / secondary peak ratio is less than 1.8 or the kurtosis... When the value is below 3, the δτ will be marked as low confidence and removed from the list. Removed from calculations, while also considering bandwidth, window length, α, threshold The specific values ​​and versions are written into the black box, and any change to any parameter must be verified through a holdout set regression before it can take effect.

[0045] Sensitivity lag triggers single-path retention and conservative weighting. In the fusion unit, the weight of the hot path is reduced and the process noise is tightened. The process is then restored in segments after the net imbalance reaches the target continuously.

[0046] Among them, the weights and process noise update law, Hot path weights With process noise according to, ; in 0.50 0.80, 0.15 <0.30; when the net imbalance of the bridge arm is continuous Each sample satisfies the threshold. At that time, restore in segments. .

[0047] Then the parameter { The source and version are written into the black box.

[0048] in, and The engineering definition is specified as follows: based on the transducer drive frequency Define carrier period and with sampling rate Label sampling period Under the default configuration, the stability threshold is determined. Balance count Take the number of samples corresponding to 5 carrier cycles. The imbalance threshold is taken as Therefore, when the sliding window W=[t− Continuous within t] Each sample satisfies The time is recorded as the balance is achieved; if it occurs at any time... The counter is immediately reset to zero and recorded as unbalanced; At the same time, to accommodate differences in media and cabling, it is permissible to include [the following] during the calibration phase. at {3,5,7}·ceil Three configuration options will Optimize within the compliance range of [0.5mV, 1.5mV] (default 1mV), and write all values, sources, version numbers and calibration sample sizes into the black box; When judged to be in a balanced and compliant state Only then is it allowed to press The recovery caliber is updated at most once every Δt, otherwise if it is continuously updated... If the imbalance is triggered multiple times, the above update will be frozen for 1 second and the settings will be reduced. But not less than Simultaneously record the trigger timestamp, The current shift parameter version is provided for traceability.

[0049] Furthermore, the behavior tags are classified based on a set of rules, and the classification results are defined as follows: If the event needle sequence shows a timing pattern of US arriving first, followed by Δp, and then T or P, and there is no repeated write-back during interval switching, it is marked as acoustic first arrival type; if Δp shows a weak breathing wave in the low differential pressure region and Re transition zone while the density interval switching rate increases, it is marked as cross-region jitter type; if the correlation lag of TH relative to P exceeds the threshold, and the net imbalance only converges at the end of the event, it is marked as viscosity lag type; if compensation freeze and version write-back re-enter within the same short window, it is marked as feedback re-entry type.

[0050] After clarifying the classification labels and source paths, continuous events are divided into different event sequence groups by combining channel type, event label category, and event timeline sequence data. Each group contains only a set of events with consistent temporal patterns, ensuring consistent semantic path convergence and continuous and stable contextual conditions. Common features such as the mean error window degree ε, the mean consistency kernel κ, interval switching frequency, freeze trigger rate, and fusion branch switching rate are extracted for each sequence group, and a feature vector set is established through sliding aggregation.

[0051] Specifically, let the set of event timestamps within the same event window be { }, c∈{ultrasound, differential pressure, temperature / pressure}, Window degree, ε= (| -median({ })|) / W, where W is the current event window width; Consistent core, Z represents the peak value of the standardized cross-correlation between channels (0~1), and Z represents the number of channels.

[0052] Qualification criteria The threshold is derived from type testing or standard caliber, and then written into the black box along with the version.

[0053] Subsequently, stability scores are calculated based on the aforementioned feature vectors, and event sequence groups are sorted according to the scores. Those with higher scores are given priority in being selected as the dominant path and target reference during the compensation / fusion process; for groups with lower scores but lower conflict levels, flexible registration and small-step weight ramping strategies are employed.

[0054] The stability score is defined as follows: Let normalized window size The conflict degree χ is obtained by weighting and normalizing the freeze trigger rate, fusion branch switching rate, and interval switching frequency. ; Stability rating ; Then the weights { The threshold source and version are written into the black box.

[0055] The weight allocation process involves first constructing training and holdout sets using typical operating condition data from the past three years, categorized by medium / installation / operating condition, and then using the analytic hierarchy process (AHP) to provide initial vectors. Under the constraint that 'weights are non-negative and the sum of the three is 1', an optimization with monotonicity constraints is performed sequentially with the objective of 'maximizing the hit rate of the dominant path and minimizing the false positive rate' on the holdout set, and default values ​​are given. =0.35、 The allowable intervals are 0.25~0.45, 0.35~0.55, and 0.15~0.35, respectively, followed by the coefficients of the conflict degree decomposition term. Joint fitting on the hold-out set using least squares and robust constraints yields default values. The permissible ranges are 0.30~0.60, 0.20~0.40, and 0.10~0.30, respectively, and the sum of the three is 1; Subsequently, the upper limit of the misaligned window. Lower limit of kernel consistency Conflict Limit The initial values ​​are jointly determined by the type test quantiles and the ROC inflection points of the holdout set, with the default value being... =0.15、 =0.65、 =0.30 (allowable intervals are 0.10~0.20, 0.60~0.75, and 0.20~0.40 respectively), when the window error rate is simultaneously satisfied within the continuous monitoring window ≤ Nuclear consistency ≥ Conflict level ≤ And maintaining this state for at least 10 seconds is considered highly stable; any instance of exceeding the limit will trigger the downgrade or freeze criteria you specified in your document; Specifically, in the compensation and fusion execution phase, mapping is not directly implemented based on a fixed priority. Instead, the conflict mapping cluster derived from the rule generation channel is referenced synchronously to dynamically calibrate and adjust the strategy for the current event path. When a high-conflict trajectory is hit, the execution strategy is reversed and compressed, pausing the regular weighted fusion and switching to a conservative output based on the dominant path and historical smoothing. Label dilution and timing gap protection are implemented for the conflict path to prevent pseudo-transients from being written into the parameter update branch.

[0056] Then, for different measurement links, the preprocessing path is differentiated and converged. In the ultrasonic forward / backward channel, envelope extraction is used to suppress isolated spikes, peak latching is used to stabilize the first arrival edge, and window smoothing is used to maintain the temporal structure, thus obtaining the first arrival time with stable morphology. In the differential pressure channel, sliding window regression is performed in the static pressure section and the zero-point residual is calculated, thereby obtaining the zero-point stability criterion and drift rate label. In the thermal bridge channel, the bridge arm ratio and reference difference are statistically tested to obtain the balanced / unbalanced state and write it into the preprocessing record with a timestamp.

[0057] If the preprocessing results show that the zero-point drift exceeds the threshold or the bridge arm is unbalanced, the cycle is placed into a restricted computation state. By setting alarms and freezing write channel operations, any coefficient update branches are temporarily suspended, and only the basic measurement link output is retained to maintain external continuity. Simultaneously, a historical output loop buffer and trigger counter are established on the embedded end, recording the filter type and order, pass / stopband settings, bias removal method, and parameter version number used, providing a chain of evidence for subsequent problem tracing and version review. This restricted state is automatically lifted after the stability criterion is continuously met, and the computation chain returns to its normal path.

[0058] Furthermore, under the premise of clean and controlled input anomalies, two fusion calibration paths are activated in parallel. External disturbances are transformed into calculable compensation quantities using clearly defined sources and rules, and algorithmic uncertainty is transformed into controllable decision outputs. Specifically, One approach is a multi-sensor signal fusion compensation path. When fluctuations in temperature, pressure, or density may introduce systematic errors, the system compensates for physical quantities through table lookup / interpolation and model evaluation. Specifically, the jet oscillation water meter obtains the compensation coefficient corresponding to the current temperature by locating the interpolation interval in the temperature and compensation coefficient table, selecting the interpolation method, and backfilling the version and interval information. Then, the oscillation frequency is injected into the correction model to obtain the instantaneous flow rate after temperature compensation, suppressing the temperature, frequency, and flow rate mapping offset from the source. The differential pressure path obtains the mass flow rate under pressure / temperature fluctuations by synchronously reading differential pressure / static pressure / temperature and calculating or looking up the density. Then, the mapping is corrected according to a predetermined order of differential pressure square root linearization, density normalization, and range coefficient correction. The thermal mass path performs online recalibration of the bridge output through the pressure and sensitivity mapping provided by the built-in pressure channel to obtain the pressure-compensated thermal output.

[0059] The power consumption / sensitivity mapping for the thermal mass path is as follows: Assume heating power consumption Bridge arm temperature difference ΔT, density ρ, and viscosity μ Pressure and sensitivity correction ; Online correction, ; Records, G caliber source, μ(T,ρ) lookup table interval and version, Sources and versions.

[0060] Meanwhile, to avoid excessive following of compensation under rapid disturbances, the above path stores the version of the coefficients used, the interpolation range and the triggering reason for each trigger; and when the rate of change of the compensation input exceeds the narrowband hysteresis criterion, it first enters a short-term freeze window to obtain a verifiable conclusion that is delayed by one update or updated immediately, thus eliminating oscillations.

[0061] Narrow band hysteresis and freezing aperture are as follows: Assume compensation input rate of change ; Enter freeze state. And continuous The sampling period is from the establishment to the freezing of M sampling cycles; Unfreeze And continuous Secondary establishment; constraint. .

[0062] The source and version are written into the black box.

[0063] Secondly, there is a multi-model / multi-algorithm fusion calibration path. When the signal morphology or noise background reduces the reliability of a single path, the system runs multiple decoding methods in parallel within the same measurement cycle and adaptively fuses them. Taking ultrasound as an example, two sets of flow results, Q1 and Q2, are obtained by independently solving the operation through cross-correlation and threshold detection methods. Then, they are compared with the historical output Q0. By calculating |Q1−Q0|, |Q2−Q0|, and |Q1−Q2| and referring to the operating condition threshold table, the branch determination is obtained. When the two paths are consistent and stable, they enter Kalman filtering for weighted fusion to obtain the fused output and covariance. The minimum skeleton and shielding inequality of the Kalman filter are as follows. state, ; Measurement, ; Gain, ; renew

[0064] Shielding inequalities or Path to shielding deviation exceeding threshold and write the shielding reason; The source and version are written into the black box.

[0065] Among them, the state variables are volumetric flow rate, temperature, pressure, and sound velocity correction terms, with the current working condition as the linearization base point. The sensitivity coefficient is calculated by mapping the density-temperature-pressure relationship with the ultrasonic arrival time difference, and then the state recursion matrix is ​​given by first-order approximation. The observations include two flow rate solutions and direct measurements of temperature and pressure. The observation matrix is ​​established by calculating and filling in the base point in the order of the partial derivatives of each with respect to the state variables. Then, the recursion is performed with a fixed sampling step size, and the linearization base point and sensitivity coefficient are refreshed once when the static section verification is passed or the working condition transition is obvious. If alignment break or cross-correlation main peak degradation is detected, the refresh is temporarily suspended and the previous effective coefficient is maintained. At the same time, the base point, sensitivity and refresh reason are written into the black box for traceability.

[0066] The values ​​of q and r are as follows: First, calculate the Allan variance of each channel during the zero-point static segment to obtain the lower bound of the random walk, and then convert it into the lower bound of the process noise. Based on this, the initial value of each state component is generally taken as q = 1.5 to 2.5 times. The starting point of the measurement noise r is then calculated based on the standard deviation of the sensor calibration certificate and the statistical synthesis of residuals during operation. Adaptive fine-tuning is then implemented using standardized residual thresholds: if the standardized residual is too large within a continuous window, the corresponding component of q or r is amplified proportionally; if it is consistently too small, it is contracted proportionally. The upper and lower limits are given by the Allan variance of the static segment and the results of field recharge, respectively. The timestamp, magnification, upper and lower limits, and version of each adjustment are written into the black box. When the temperature and pressure enter the cross-phase region or the observation degrades, r is locked, q is relaxed, and the recalculation period is shortened. After stabilization, the original period is used for regression.

[0067] The parallel solution using two methods for ultrasound ToF is as follows. Let the effective sound path length L and the included angle θ be the arrival times with and against the current. , Δt= - Threshold method ; Cross-correlation method

[0068] Write: Method used, pre-filter settings, correlation window length, Version and range information are uniformly incorporated into the fusion and discrimination.

[0069] If one path deviates beyond the threshold, that path is blocked and its weights are redistributed. Only reliable paths are used in conjunction with historical values ​​for smoothing, resulting in a conservative and stable output. If both paths are unstable but the differences are small, a short-term hold condition is set and an unlock condition is established to maintain controlled output switching during recovery. The same logic is used for differential pressure and hot links. Each judgment is recorded with the branch used, weights, blocked path and reason for blocking, source of Kalman parameters, covariance initialization and reset conditions, all of which are included in the fusion trajectory for post-event review.

[0070] To ensure the time difference directly translates into volumetric flow rate, the arrival time difference between downstream and upstream flows is denoted as the time difference quantity. The equivalent sound path within the pipe is L, the angle with the pipe axis is the installation angle, the pipe cross-sectional area is A, and geometric and installation errors are included in the correction coefficient. Therefore, under the condition of low flow velocity relative to sound velocity, the theoretical volumetric flow rate is: =Correlation coefficient·A·(Time difference·Sound speed²) / (2·L·cos installation angle), the cross-correlation method substitutes the time difference of the correlation peak into the same mapping to obtain = Correction coefficient·A·(Related peak time difference·Sound velocity²) / (2·L·cos installation angle); The sound velocity is corrected by linear temperature and pressure by default, and the approved medium-specific table is used first. That is, sound velocity = reference sound velocity + temperature coefficient × (current temperature - reference temperature) + pressure coefficient · (current pressure - reference pressure). When it enters the cross-phase or table coverage area, the value is changed to be looked up in the table and extrapolation is prohibited. The temperature coefficient and pressure coefficient are derived from the joint regression of type test / field reinjection and the range and version are written to the black box. If the temperature and pressure exceed the limit, it will fall back to the table value and record the reason for the fallback and the timestamp.

[0071] Then, for differential pressure and thermal measurements, calibration is performed through parallel solving, difference discrimination, and adaptive fusion / shielding process consistent with the ultrasonic channel. Specifically, in the differential pressure channel, two sets of solvers, differential pressure square root linearization path and density normalization regression path, are run simultaneously and time-scaled with a unified hardware timer to obtain two sets of instantaneous flow solutions QΔp,1 and QΔp,2. Next, within the thermal channel, by simultaneously operating two sets of solvers—a differential bridge, a power consumption estimation path, and a pressure and sensitivity recalibration path—two sets of instantaneous flow solutions, QT,1 and QT,2, are obtained.

[0072] Subsequently, the dual-path solution of each channel is compared with the historical output Q0 of the previous cycle using difference measures |Q1−Q0|, |Q2−Q0|, and |Q1−Q2|, and the operating condition threshold table to obtain branch discrimination. When the two paths are consistent and stable, the results of the two paths are weighted and fused using Kalman filtering to obtain the fused output and covariance. When one path deviates beyond the threshold, the path is masked and the weights are redistributed, and only the reliable path combined with historical values ​​is used for smoothing to obtain a conservative and stable output. When both paths are unstable but the difference is not significant, a traceable process of holding and recovering is obtained through short-term hold and unlock condition constraint operations.

[0073] Differential pressure metering mapping is defined as follows: Differential pressure mass flow rate is implemented according to one of the following specifications, and the version and source are specified. Standard caliber, QΔp(t) = ; Simplified, QΔp(t) = ; Where A is the effective area of ​​the throttling element; C is the range segmentation coefficient; C(Re) is the Reynolds correction.

[0074] Each branch discrimination involves writing down the branch used, weight, blocked path and blocking reason, Kalman parameter source, covariance initialization and reset conditions to obtain a timestamped record, which is then uniformly written into the fusion trajectory for review.

[0075] Subsequently, the closed-loop application was completed by introducing the fused and corrected flow value into two implementation mechanisms: calibration control output and parameter self-update. On the one hand, at the end of each measurement cycle, the online compensation output directly replaces the uncompensated value with the corrected value, completes data frame encapsulation, register refresh, communication transmission and readback operations, and obtains a consistent display and real-time output that is transparent to the upper system through the digital communication port without manual intervention. To ensure consistency, each frame of output is processed by adding a timestamp, device identifier, channel number, fusion branch and weight summary, and the version of the coefficients used to obtain a traceable output message.

[0076] On the other hand, when any of the following conditions are met—long-term deviation, abnormal drift rate, or decreased confidence—the offline calibration coefficient update performs historical sample screening and segmented fitting, residual evaluation, version writing, verification, and version switching operations during short load cycles or maintenance periods to obtain effective parameters with a new version number and verification mark. During version switching, the old version is retained and the index is switched simultaneously to obtain a version chain that can be rolled back at any time.

[0077] Both mechanisms, when triggered, obtain a complete chain of evidence that can be made available to the outside world by writing timestamps, snapshots of input parameters, compensation / fusion branches and versions of coefficients used, triggering reasons and result status to black box records and read-only diagnostic registers.

[0078] Meanwhile, to prevent data or rules from unknown sources from entering the decision-making process, guardrails are continuously implemented through three types of pre-verification during operation. When it is detected that any channel is missing a three-element binding, any compensation table is missing a version identifier or fails verification, or any fusion threshold is not calibrated and confirmed, the corresponding compensation / fusion branch is immediately stopped and set to a restricted state, the branch is circuit-broken, the last valid output is maintained, and the abnormal entries are entered into the database, so as to obtain stable output and verifiable alarms. If similar alarms occur consecutively within a specified sliding window, marking the branch as pending review and retaining only the basic measurement link output and timestamp operations achieves the isolation effect of preventing unexplained data from participating in calibration. If version write verification fails or the number of times exceeding the limit after an update reaches a threshold, rollback control immediately switches to the previous available version and generates a rollback record, resulting in a standardized rollback chain.

[0079] In scenarios where three conditions simultaneously hold true: rapid temperature jumps, moderate pressure fluctuations, and density model switching within critical intervals, using fixed compensation coefficients and fixed fusion weights often leads to concurrent phenomena of compensation lag and frequent branch switching. To address this, a strategy of delaying compensation by one step and maintaining weights during the disturbance period is first developed on the compensation side by establishing narrow-band hysteresis criteria for temperature and pressure and loading a short-term freeze window operation upon triggering. Then, on the fusion side, a strategy of restoring weighted output based on operating conditions during the recovery period is obtained through differential convergence detection and weight recovery operations. Furthermore, in scenarios with reduced signal-to-noise ratio at low flow rates, transducer output attenuation, and fluid entrainment with microbubbles, a stable output suppressing false triggers and small signal jitter is achieved by increasing the trigger threshold, extending the sliding window, and enabling a single-path retention strategy with cross-correlation as the primary factor and threshold as the secondary factor. In scenarios with concurrent scenarios of pipeline mechanical vibration, increased power supply ripple, and increased electromagnetic interference, a balanced approach is achieved by temporarily increasing the filtering order, tightening the synchronization window, and reducing process noise during the disturbance period, ensuring neither insufficient filtering nor excessive phase lag.

[0080] This invention also relates to an embedded multi-parameter fusion calibration control system for flow meters, such as... Figure 1 As shown, it includes: a data acquisition and density calculation module, an event classification and scoring module, a convergence and strategy execution module, and a data storage module, with signal connections between the modules; The data storage module is used to store all data during the storage process; The acquisition and density calculation module is used to sequentially activate the primary flow measurement signal and environmental parameter channels, set the communication and pin positions and write the three-element binding, and sample differential pressure, forward and reverse flow first arrival, bridge differential judgment bridge arm balance, and acquire temperature and pressure to calculate density according to two diameters; The event classification and scoring module is used to extract frame information of the clean sequence of each channel, monitor and mark alignment breaks, compensate for feedback re-entry, sensitivity hysteresis and generate rule sets; after the classification labels and source paths are clear, the channels are grouped by channel type, event labels and time order and stability scores are calculated to select the dominant path;

[0081] The convergence and strategy execution module is used to perform differentiated convergence on different measurement link preprocessing paths, determine zero drift exceeding the threshold or bridge arm imbalance to set the computation state and freeze the write channel, and after the anomaly is under control, the compensation path and multi-algorithm path are opened in parallel.

[0082] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0083] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0084] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0085] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0086] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0087] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An embedded multi-parameter fusion calibration and control method for flow meters, characterized in that: include: Activate the flow measurement signal and environmental parameter channels in sequence, set the communication and pin positions and write the three-element binding, and sample differential pressure, forward and reverse flow first arrival, bridge differential judgment bridge arm balance, and collect temperature and pressure to calculate density according to two diameters; Extract frame information of clean sequences for each channel, monitor and mark alignment breaks, compensate for feedback re-entry, and sensitivity hysteresis, and generate a set of rules; after the classification labels and source paths are clear, group them according to channel type, event label and time order, calculate stability scores and select the dominant path; Differentiated convergence is applied to the preprocessing paths of different measurement links. If the zero drift exceeds the threshold or the bridge arm is unbalanced, the calculation state is restricted and the write channel is frozen. After the anomaly is under control, the compensation path and the multi-algorithm path are opened in parallel.

2. The embedded multi-parameter fusion calibration control method for flow meters according to claim 1, characterized in that: Activate the flow measurement signal channel and environmental parameter channel in sequence; set communication parameters and interface pins according to the data channel registration table, write the timestamp, device identifier and channel number three-element binding, load the temperature compensation coefficient, density correction relationship and range segment correction curve, and perform port level and pull-up or pull-down checks, communication handshake detection and register loopback verification.

3. The embedded multi-parameter fusion calibration control method for flow meters according to claim 2, characterized in that: Sampling differential pressure digital samples, first arrival times of forward and reverse currents, and bridge differential signals to determine bridge arm balance; Temperature and pressure are collected by a thermistor and a pressure transmitter, and density is calculated according to the state equation or calibration curve. Then, the system polls at a uniform sampling period and performs noise reduction and blanking in a ring buffer.

4. The embedded multi-parameter fusion calibration control method for flow meters according to claim 1, characterized in that: After the preprocessing results are produced, the source identifier, timestamp, event pin index, filter version and sampling phase, lookup table or model interval and version of each clean sequence are extracted and written into the black box; then, three types of offsets are monitored and marked: alignment break, compensation feedback reentry, and sensitivity hysteresis; when the offsets coincide or persist, a set of rules is generated accordingly.

5. The embedded multi-parameter fusion calibration control method for flow meters according to claim 4, characterized in that: The set of rules is as follows: Relative timescales are reconstructed using ultrasound first arrival or differential pressure second-order zero crossover as needle points and delayed compensation injection is performed. Density back-writing is delayed once in the critical region and temporary density proxy is generated by using arrival time difference / envelope half-width difference / single point temperature as observation quantity. After backfilling the record, thermal path weight is adjusted in the fusion unit and process noise is tightened. After the criteria are met, segmented recovery is performed, and parameter source and version are recorded simultaneously. Then, according to the rule set, the classification and labeling are performed as acoustic first type, cross-regional jitter type, viscosity hysteresis type, and feedback re-entry type.

6. The embedded multi-parameter fusion calibration control method for flow meters according to claim 5, characterized in that: After clarifying the classification labels and source paths, the event sequence groups are divided according to channel type, event label, and time order. For each group, the mean error window, mean consistency kernel, interval switching frequency, freeze trigger rate, and fusion branch switching rate are extracted and aggregated to generate feature vectors. Then, the stability score is calculated and sorted based on the feature vectors. The high-scoring group is selected as the dominant path and target alignment reference, while the low-scoring group with low conflict degree is subjected to flexible registration and small-step weight climbing. Simultaneously, the synchronous reference conflict mapping cluster implements dynamic calibration and strategy adjustment. When a high-conflict trajectory is hit, it performs strategy reverse compression, suspends regular weighted fusion, adopts a conservative output of dominant path and historical smoothing, and implements label dilution and time gap protection for conflict paths.

7. The embedded multi-parameter fusion calibration control method for flow meters according to claim 1, characterized in that: Envelope extraction, peak latching and window smoothing are performed on the ultrasound channel; zero-point residuals are calculated by sliding window regression in the static pressure section of the differential pressure channel; the bridge arm ratio and reference difference are statistically tested and time-stamped for the thermal bridge channel; and zero-point and equilibrium state labels are generated for event classification and stability scoring. When the preprocessing determines that the zero-point drift exceeds the threshold or the bridge arm is unbalanced, a restricted computing state is set, the write channel is frozen, and only the basic output is retained.

8. The embedded multi-parameter fusion calibration control method for flow meters according to claim 7, characterized in that: Two fusion calibration paths are activated in parallel under abnormally controlled conditions: The compensation path is based on table lookup or model injection of temperature, pressure, and density compensation, and the version and triggering reason are recorded. The multi-algorithm path runs dual solvers in each channel, compares with historical output and operating condition thresholds, performs fusion or masking, and records branches, weights and masking reasons, and writes them into the fusion trajectory to support sequence grouping, stability scoring and conflict mapping.

9. An embedded multi-parameter fusion calibration control system for flow meters, characterized in that: include: The module includes a data acquisition and density calculation module, an event classification and scoring module, a convergence and strategy execution module, and a data storage module, with signal connections between the modules. The data storage module is used to store all data during the storage process; The acquisition and density calculation module is used to sequentially activate the primary flow measurement signal and environmental parameter channels, set the communication and pin positions and write the three-element binding, and sample differential pressure, forward and reverse flow first arrival, bridge differential judgment bridge arm balance, and acquire temperature and pressure to calculate density according to two diameters; The event classification and scoring module is used to extract frame information of the clean sequence of each channel, monitor and mark alignment breaks, compensate for feedback re-entry, sensitivity hysteresis and generate rule sets; after the classification labels and source paths are clear, the channels are grouped by channel type, event labels and time order and stability scores are calculated to select the dominant path; The convergence and strategy execution module is used to perform differentiated convergence on different measurement link preprocessing paths, determine zero drift exceeding the threshold or bridge arm imbalance to set the computation state and freeze the write channel, and after the anomaly is under control, the compensation path and multi-algorithm path are opened in parallel.

Images