A smart weighing device based on multi-cycle continuous judgment and its online fault diagnosis and self-healing method

Abstract

This invention belongs to the field of industrial weighing and measurement technology, and discloses an intelligent weighing device based on multi-cycle continuous judgment and its online fault diagnosis and self-healing method. The device includes a microcontroller configured to: extract sensor signal feature values ​​at preset cycles and compare them with no-load reference values; mark anomalies when values ​​exceed limits; count the number of consecutive anomalies through a sliding window; determine permanent faults and trigger self-healing when a preset number of anomalies is reached; otherwise, determine transient interference. The device also measures the no-load values ​​of each channel in an empty weighing state; when the same value exceeds the threshold multiple times consecutively, it determines unidirectional drift and corrects the reference value. The self-healing process includes temporary compensation and early warning for slight deviations, and removal of faulty channels for long-term threshold exceedances. The device also distinguishes between short-term fluctuations, material imbalance, and structural faults by the number of consecutive detections, and performs online self-correction for material imbalance. This invention accurately distinguishes between interference and faults, adaptively adjusts thresholds, automatically maintains weighing after a fault, and is low-cost and easy to deploy.

Description

Technical Field

[0001] This invention belongs to the field of industrial weighing and measurement technology, specifically relating to an intelligent weighing device and its control method that can perform online monitoring, fault classification, adaptive threshold adjustment, graded self-healing, off-center load deviation diagnosis, mechanical structure abnormality early warning, and sensor remaining life prediction during real-time weighing. It is particularly suitable for continuous operation scenarios such as belt scales, hopper scales, and batching systems. Background Technology

[0002] Load cells are core components of industrial batching and metering systems. In actual industrial settings, these sensors operate in harsh environments for extended periods, leading to frequent malfunctions. Particularly challenging is the highly concealed nature of localized faults in individual sensors, making it impossible for existing weighing devices to independently monitor each sensor. Furthermore, factors such as weighing frame deformation, uneven material loading, belt misalignment, electromagnetic interference, and temperature drift can all contribute to inaccurate weighing data.

[0003] Currently, conventional weighing devices have the following common defects: First, they can only detect overall continuity and cannot locate local faults in individual sensors; second, they use fixed deviation thresholds and cannot adapt to individual differences and dynamic changes; third, they are prone to misjudging instantaneous electromagnetic interference as sensor faults, resulting in frequent false alarms.

[0004] More importantly, there is a long-standing technical bias in this field: the widespread belief that sensor malfunctions must trigger an immediate alarm, and any delay will lead to increased scrap. Therefore, existing solutions all adopt the logic of "determining a fault upon a single anomaly," resulting in the bias of "better to have a thousand false alarms than one missed alarm." However, the actual pain point in industrial settings is precisely that too many false alarms lead operators to disable alarm functions, ultimately missing the real fault.

[0005] While imported high-end devices possess diagnostic capabilities, they are expensive and complex. Existing patents (such as CN201611276U) disclose a multi-channel independent A / D conversion hardware architecture, but these are only used for signal display during the installation and commissioning phase and lack online monitoring and self-healing capabilities. Other solutions rely on cloud servers and neural network algorithms, which are costly and lack real-time performance. Summary of the Invention

[0006] To address the above problems, the present invention aims to provide a low-cost intelligent weighing device that is easy to deploy, can accurately distinguish between interference and faults online, adaptively adjusts thresholds, and automatically maintains weighing after a fault.

[0007] To achieve the above object, the present invention provides the following technical solutions: An intelligent weighing device based on multi-cycle continuous judgment and its online fault diagnosis and self-healing method, including a device housing, a plurality of sensor input interfaces provided on the housing, M independent filtering and amplification circuits provided in the housing, an M-channel A / D conversion chip, a single-chip microcomputer, a built-in temperature sensor, a display and alarm unit, a data storage unit, and a communication interface. The device further includes a signal validity verification function for detecting abnormal states of non-weight deviation types such as abnormal excitation voltage and continuous non-output of effective weight signals in each channel sensor. When such an abnormality is detected, even if the current weight value is within the allowable deviation range, the system still marks this channel as abnormal, incorporates it into the subsequent fault judgment process, and is configured to execute the following core control algorithm.

[0008] The present invention adopts a "multi-cycle continuous judgment" delay confirmation mechanism instead of the immediate judgment mechanism commonly used in the art. The mechanism is as follows: Let the weight value of the i-th sensor at the discrete sampling time t be x_i(t), its current no-load reference value be μ_i(t), and the lower and upper limits of the allowable deviation be Δ_i⁻ and Δ_i⁺ respectively. Let FS be the full-scale value of the sensor. The single-chip microcomputer collects the weight data of each channel at a sampling rate of 512Hz for real-time weighing and jitter analysis. At the same time, to filter out instantaneous electromagnetic interference, the fault judgment uses an independent downsampling period: One characteristic value is extracted from the high-speed sampling data every 0.5 seconds as the input of the fault detection period. Define the detection period T_d = 0.5 seconds. When the continuous detection times L = 50, the total fault confirmation duration is L×T_d = 25 seconds.

[0009] Define the abnormal flag function f_i(t): When the deviation of the detected value from μ_i(t) exceeds the allowable deviation range [Δ_i⁻,Δ_i⁺], f_i(t)=1; otherwise f_i(t)=0. The system maintains a sliding window with a length of L (L≥50). By accumulating all the abnormal flag values within the window, the continuous abnormal times S_i(t) are statistically obtained:

[0010] S_i(t) = ∑_{k=0}^{L-1} f_i(t-k)

[0011] where S_i(t) is the continuous abnormal times of the i-th sensor at time t, L is the preset length of the sliding window, and f_i(t-k) is the abnormal flag function value of the i-th sensor at time (t-k). When S_i(t)=L, that is, abnormal conditions occur in L consecutive detection periods, it is determined that the i-th sensor has a permanent hardware failure, and the alarm and self-healing process is triggered; when S_i(t)<L, it is determined as instantaneous electromagnetic interference, no alarm is triggered, and the counter is automatically cleared in the subsequent normal period.

[0012] An adaptive reference value generation and dynamic maintenance algorithm was established. After the weighing body is calibrated, the average AD value of N sampling periods under empty weighing conditions is recorded as the initial empty reference value μ_i(0):

[0013] μ_i(0) = 1 / N ∑_{j=1}^{N} x_i(t_j)

[0014] Where μ_i(0) is the initial no-load reference value of the i-th sensor, N is the number of sampling periods, and x_i(t_j) is the weight value of the i-th sensor in the j-th sampling period. This initial no-load reference value is stored in non-volatile memory. Simultaneously, the system automatically sets an allowable deviation range [Δ_i⁻, Δ_i⁺] for each sensor based on the field conditions; this range can also be adjusted by the user.

[0015] Each time power is supplied, the device loads the last reference value μ_i stored before the last power outage, measures the current no-load average value x̄_i, and calculates the relative deviation δ_i:

[0016] δ_i = ( | x̄_i - μ_i | ) / FS × 100%

[0017] Where δ_i is the relative deviation of the i-th sensor, x̄_i is the currently measured average value under no-load conditions, μ_i is the stored no-load reference value, and FS is the full-scale value of the sensor. If δ_i ≤ 2%, the original reference value is maintained; if δ_i > 2%, the system does not correct immediately but waits for a complete loading and unloading cycle (empty scale → loading → unloading back to empty scale). After unloading, the average value of the empty scale x̄_i' is measured again. If |x̄_i' - x̄_i| ≤ 0.1%FS, the scale is confirmed to be in normal condition, the current no-load value is updated to the new reference value, and the stored μ_i←x̄_i' is overwritten. If they are inconsistent, it is determined that there is residual material or mechanical jamming in the scale, and an "abnormal scale condition" warning is issued.

[0018] During long-term continuous operation, the device continuously monitors the changing trend of the no-load value of each channel. The drift detection window length is defined as M=5. When all sampling points within the window satisfy x_i>μ_i+0.2%FS, compensation correction is initiated, and the no-load reference value μ_i is updated.

[0019] μ_i ← μ_i + η · (1 / M) ∑{j=1}^{M} ( x{i,j} - μ_i )

[0020] Where η is the adaptive learning rate, with a value range of 0 < η ≤ 1; M is the drift detection window length; and x_{i,j} is the unloaded value of the j-th sampling point within the window. This compensation correction only applies to "continuous unidirectional drift" caused by factors such as material accumulation and aging. For "bidirectional fluctuations" caused by temperature changes, since they do not meet the "continuous unidirectional" condition, the automatic zero-point drift compensation algorithm of this invention will not produce any response. This is fundamentally different from the conventional zero-point tracking method in the prior art that indiscriminately tracks and compensates for all drifts.

[0021] The microcontroller also monitors the deviation d_i of the current no-load value and the initial no-load reference value of each channel in real time.

[0022] d_i = ( | μ_i(t) - μ_i(0) | ) / FS × 100%

[0023] Where d_i is the deviation of the i-th sensor, μ_i(t) is the current no-load reference value, μ_i(0) is the initial no-load reference value, and FS is the full-scale value. When the deviation d_i of a certain channel is greater than 0.5% and other channels are normal, it is determined that the sensor corresponding to that channel may have zero-point drift, moisture, or mechanical jamming faults, and a corresponding warning is issued.

[0024] Short-term impact identification function has been implemented. Impact intensity p_i is defined as follows:

[0025] p_i = ( x_i^peak - μ_i ) / FS × 100%

[0026] Where p_i is the impact intensity of the i-th sensor, x_i^peak is the maximum weight value of the i-th sensor within 0.5 seconds, μ_i is the current no-load reference value, and FS is the full-scale value. When p_i > 30%, the system enters the impact confirmation process: waiting for the empty scale state after unloading is completed, and measuring the empty scale value x_i^after. If |x_i^after - μ_i| < 0.2%FS, it is determined to be a temporary heavy-load impact, and no fault alarm is triggered; if |x_i^after - μ_i| ≥ 0.2%FS, it is determined to be permanent sensor damage or mechanical jamming, and a fault alarm is immediately triggered and the system enters self-healing mode. In addition, even if the impact intensity p_i ≤ 30%, if the zero return deviation after multiple consecutive empty scale cycles continues to exceed 0.2%FS, it should be determined to be sensor zero drift or mechanical jamming, and a corresponding warning should be issued.

[0027] An adaptive threshold method for detecting jitter anomalies in single-sensor signals was implemented. The microcontroller acquires weight data from each channel at a high-speed sampling rate of 512Hz, with a window length W=50, and calculates the jitter amplitude D_i.

[0028] D_i = max_{k=1}^{W} x_i(k) - min_{k=1}^{W} x_i(k)

[0029] Where D_i represents the jitter amplitude of the i-th sensor, x_i(k) is the weight value of the k-th sampling point within the window, and W is the window length. The system calculates the jitter amplitude of all channels in real time and maintains a long-term sliding window to obtain the overall average jitter amplitude D̄_all. When the jitter amplitude D_i of a certain channel exceeds 1.5×D̄_all for three consecutive time windows, a level one warning is issued; when it exceeds 4×D̄_all for five consecutive time windows, a level two alarm is issued and a self-healing mode is triggered. This adaptive mechanism can adapt to the vibration levels of different field environments and avoid misjudgments caused by fixed thresholds.

[0030] To improve the accuracy of jitter recognition, this invention extends the multi-dimensional feature jitter recognition function. The microcontroller collects weight data from each channel in real time at a sampling rate of 512Hz, and calculates the jitter amplitude D_i, jitter frequency f_i=N_zero / T_w, waveform factor W_f=X_rms / X_avg, skewness coefficient S_k=1 / W∑((x_i(k)-x̄_i) / σ_i)³, and kurtosis coefficient K_u=1 / W∑((x_i(k)-x̄_i) / σ_i) within a fixed time window (W=50). 4 -3 Five-dimensional feature parameters. The system has a preset jitter feature library, which includes feature parameter ranges for four typical fault modes: loose wiring, electromagnetic interference, mechanical vibration, and sensor damage. The microcontroller compares the extracted feature vectors with the feature library and uses a weighted similarity matching method to identify the jitter type. Confirmation is achieved after three consecutive windows of consistent identification results, and a graded warning is output based on the identification results. While each of the above feature parameters is individually known in the field of signal processing, applying these parameters in a specific combination to classify and identify the causes of jitter in weighing sensors is a first in the field of weighing and metrology.

[0031] A multi-sensor data consistency self-verification algorithm was established. Under empty scale conditions, the algorithm determines whether there is a sensor malfunction by comparing the consistency of the weight values ​​distribution across each channel. Let K be the number of connected sensors, and calculate the mean x̄ = 1 / K∑x_i and the variance σ² = 1 / K∑(x_i-x̄)². When σ² exceeds a preset consistency threshold, a sensor malfunction is determined. A step-by-step elimination method is used to locate the faulty sensor: the i-th sensor is eliminated sequentially, and the variance σ_{-i}² of the remaining channels is calculated. When the variance significantly decreases after eliminating a certain channel, that channel is considered faulty. This verification method can also be used to distinguish between environmental factors and sensor malfunctions: when all channels drift synchronously and the variance does not change significantly, it is determined to be an environmental factor; when only a few channels drift, it is determined to be a sensor malfunction.

[0032] A multi-factor automatic differentiation and collaborative working logic was established. Multiple diagnostic algorithms run in parallel based on different trigger conditions and judgment logic, automatically distinguishing between internal sensor faults and external environmental factors through a hierarchical mechanism. After each loading and unloading cycle, the system automatically measures the empty scale value. If the empty scale value deviates from the historical benchmark value by more than 0.1%FS after unloading, the system enters the multi-factor analysis process. Under extreme weather conditions, temperature drift is automatically distinguished from material accumulation and aging through multi-sensor synchronization verification, duration judgment, and rate of change-assisted judgment. The built-in temperature sensor collects ambient temperature data in real time. When the multi-sensor consistency verification determines that all channels are synchronously drifting, the system automatically retrieves the temperature change curve for comparison and verification, including temperature and drift synchronization verification and temperature change rate-assisted judgment. The system also performs multi-sensor group trend analysis by calculating the correlation coefficient matrix of the empty scale values ​​of each channel and the outlier degree of the drift speed of each channel. When the correlation coefficient of a channel with other channels decreases significantly or the deviation exceeds 2 times the standard deviation, the channel is judged to be abnormal.

[0033] To ensure that multiple diagnostic algorithms operate in parallel without interfering with each other and with clear logic, this invention establishes a complete data and logic isolation mechanism within the microcontroller.

[0034] At the physical level, the microcontroller's internal memory is divided into multiple independent dedicated areas, including: an anomaly flag sliding window cache for the multi-cycle continuous judgment module (address 0x1000~0x11FF, length L=50), a drift detection window cache for the adaptive reference value maintenance module (address 0x1200~0x12FF, length M=5), a high-speed sampling window cache for the jitter detection module (address 0x1300~0x13FF, length W=50 per channel), a long-term average jitter amplitude cache for the jitter long-term statistics module (address 0x1700~0x17FF, 1000 time windows), a sampling point cache for the off-center load diagnosis module (address 0x1400~0x14FF), a daily no-load zero-point historical record for the remaining life prediction module (address 0x1500~0x15FF), and a snapshot of the original no-load reference value before compensation for the sensor force unevenness detection module (address 0x1600~0x16FF). The threshold parameters of each module are stored in different sectors of non-volatile memory and do not overwrite each other.

[0035] Each detection module has its own independent data buffer, threshold parameters, and status flags. They do not share intermediate variables and read only raw data from the same data source. Modifications to shared data (such as the unloaded baseline value μ_i(t)) are subject to strict trigger condition verification, and the amount of modification is limited by the adaptive learning rate η. After the baseline value is updated, a global flag notifies other modules to refresh the reference value in the next calculation cycle. The alarm and warning flags of each module are independent and can be output simultaneously without interference.

[0036] At the timing level, a staggered processing mechanism is adopted: A / D sampling and high-speed jitter detection are executed every 8ms (highest priority interrupt), multi-cycle continuous judgment and short-term impact identification are executed every 0.5 seconds (timer interrupt), off-center load diagnosis and off-center load self-correction are executed every 10 seconds (software timer), adaptive baseline value maintenance is executed when the unloading event is triggered, remaining life prediction is executed every hour, and multi-sensor data consistency self-verification is executed upon power-on or every 24 hours. The above memory address allocation and interrupt priority settings are based on the specific resource configuration scheme of the microcontroller hardware architecture. Through the above mechanism, this invention ensures that multiple diagnostic algorithms are logically completely independent, timing-independent, and data-free, while sharing the same physical data source.

[0037] It should be noted that the memory address allocation and peak-shifting mechanisms described above are independent of specific diagnostic algorithm types. Their core idea is to allocate an independent storage area and trigger timing for each functional module, with modules synchronizing their states only through global flags, thereby ensuring logical isolation and data security during parallel operation of multiple modules. This mechanism can be implemented independently of any specific diagnostic algorithm in various embedded multitasking systems.

[0038] It implements self-diagnostic functions for off-center loading, misalignment, and mechanical structure anomalies. The off-center loading detection cycle T_b = 10 seconds. The microcontroller calculates the mean x̄ and variance σ² of the weight values ​​for each channel, as well as the maximum and minimum channel difference Δx_max-min. Based on continuous detection results, the off-center load status is divided into the following levels: For short-term off-center load (less than 7 consecutive detections exceeding the limit), only the event is recorded and no alarm is triggered; for left and right off-center loads, within a preset evaluation time window (e.g., within 3 minutes), if the system detects alternating left and right off-center loads or an unfixed off-center load direction, it outputs a "Material off-center load, please adjust the feeding device" prompt and activates the off-center load self-correction function; when the number of consecutive detections exceeding the limit reaches 7 or more but less than 12, and the off-center load direction is fixed, it is judged as moderate off-center load, and an alarm is issued: "Belt tension may be uneven, calibration is recommended" or "Belt misalignment, please check"; for unilateral off-center load (12 or more consecutive detections exceeding the limit and the off-center load direction is fixed), it is judged as severe off-center load, and an alarm is issued: "Belt misalignment is serious, please adjust belt tension" or "Weighing frame may be deformed, stop the machine immediately for maintenance". The consecutive number threshold in this invention is specifically set based on the specific physical laws of the weighing process: less than 7 consecutive times (corresponding to 70 seconds) is used to filter short-term fluctuations; 7 to 11 consecutive times (corresponding to 70 to 110 seconds) are used to identify moderate off-center loads that need attention; the cumulative detection of off-center load events in different directions within 3 minutes (corresponding to 180 seconds) is used to identify uneven material distribution or belt misalignment; and 12 consecutive times (corresponding to 120 seconds) of off-center load in the same direction is used to confirm structural faults that require shutdown for maintenance.

[0039] The off-center load self-correction function corrects the weight value of each channel online through a dynamic compensation coefficient γ_i:

[0040] γ_i = (x̄ / x_i) · β_i

[0041] Where γ_i is the dynamic compensation coefficient of the i-th sensor, x̄ is the average weight of each channel, x_i is the current weight value of the i-th sensor, and β_i is the position weight coefficient of each channel obtained in advance through calibration by loading standard weights at the four corners. The corrected total weight value W_corrected is:

[0042] W_corrected = ∑_{i=1}^{K} γ_i · x_i

[0043] Where K is the total number of connected sensors.

[0044] It can automatically distinguish between uneven sensor force and material imbalance. Uneven sensor force detection is based on a snapshot of the original no-load reference value before compensation to avoid structural problems being masked by zero-point drift compensation. The deviation of the current no-load reference value of the i-th sensor from the initial no-load reference value is defined as d_i = (|μ_i(t) - μ_i(0)| / FS) × 100%, and the variance of the no-load reference value of each channel under no-load conditions is defined as σ_empty² = 1 / K∑(μ_i(t) - μ̄(t))². When the deviation d_i of a certain channel is > 0.5% and other channels are normal, it is judged as a single sensor force abnormality; when the no-load variance σ_empty² exceeds the preset no-load consistency threshold, it is judged as a deformation of the weighing frame structure. Material imbalance detection is based on the variance and maximum and minimum difference of the weight values ​​of each channel under weighing conditions. When alternating left and right imbalance is detected within a preset time window, a material imbalance prompt is output. When the off-center load direction is fixed for a long period of time, the system further analyzes: if the empty scale value is normal after unloading and the off-center load direction changes with the batch of materials, it is determined that the material is off-center loaded; if the empty scale value has a systematic deviation after unloading and the off-center load direction remains unchanged for a long period of time, it is determined that the sensor is under uneven force or the scale frame is deformed.

[0045] When an off-center load is detected and confirmed to be material off-center, the device automatically activates the off-center load self-correction function. During correction, the system first calculates the average weight value of all sensors, x̄=1 / K∑x_i, as a reference, and then calculates the dynamic compensation coefficient γ_i=(x̄ / x_i)·β_i for the i-th sensor, where β_i is the position weight coefficient of each channel obtained in advance by loading standard weights at the four corners. The corrected total weight of the scale is W_corrected=∑γ_i·x_i. This function is completed entirely online automatically without the need for manual loading of weights.

[0046] Different handling strategies were adopted for uneven material loading and uneven sensor stress. Uneven sensor stress stems from uneven mounting base, sensor tilt, or weighing frame deformation, which is a structural fault. For this type of fault, a "diagnostic and early warning" strategy was adopted, clearly prompting users to perform mechanical maintenance to eliminate the fault at its root. Uneven material loading, on the other hand, is caused by fluctuations in the operating status of the feeding device, resulting in uneven material distribution, which is a fluctuation in operating conditions. For this type of problem, a combined "self-correction + prompting" strategy was adopted, compensating for weight errors online to ensure production continuity, while simultaneously prompting users to adjust the feeding device within appropriate timeframes.

[0047] A fault-grading self-healing and dynamic reconfiguration mechanism has been established. When an anomaly is detected in a certain sensor, the system adopts a graded handling strategy based on the severity and development trend of the anomaly:

[0048] For minor deviations that occur in the short term, the system uses a temporary compensation method, using the detected deviation as the compensation value to correct the current reading, while issuing a first-level early warning prompt for inspection, and continuously monitoring the long-term trend of the channel.

[0049] When long-term trend analysis confirms that the offset has continued to increase and exceeds the preset threshold, or when a sensor is detected to have completely failed, the system determines that the sensor has suffered irreversible performance degradation or permanent hardware failure. At this time, the data of that channel is completely removed, the number of effective channels is reduced by 1, and an alarm of the corresponding level is issued.

[0050] When the number of faulty sensors rejected reaches the preset maximum allowable value, the device immediately stops the self-healing mode and issues an emergency shutdown alarm signal. After the rejected channel returns to normal, it can be automatically reinstated for weighing after confirmation through multiple consecutive detection cycles.

[0051] If no abnormalities are detected in a removed channel within the subsequent L_r consecutive detection cycles (L_r=100, corresponding to approximately 50 seconds), the channel will be automatically reinstated for weighing. Simultaneously, the channel's historical fault information will be stored in the data storage unit for future maintenance queries.

[0052] The function of predicting the remaining lifespan of the sensor has been implemented. The sensor performance degradation index E_i(t) is defined as follows:

[0053] E_i(t) = ( μ_i(t) - μ_i(0) ) / FS × 100%

[0054] Where E_i(t) is the performance degradation index of the i-th sensor at time t, μ_i(t) is the current no-load reference value, μ_i(0) is the initial no-load reference value, and FS is the full-scale value.

[0055] The microcontroller periodically records the no-load zero-point values ​​of each channel and uses the moving average method to calculate the average drift velocity v_i(t) over the most recent K days (K=7):

[0056] v_i(t) = 1 / K ∑_{k=0}^{K-1} ( E_i(tk) - E_i(tk-1) ) / Δt

[0057] Where v_i(t) is the average drift velocity, K is the number of days in the sliding window, and Δt is the time interval between two adjacent records.

[0058] When v_i(t) > 0.05% / day and continues for more than 7 days, the sensor performance is determined to have entered a period of accelerated degradation, an early warning signal is issued, and the remaining usable days L_rem are estimated.

[0059] L_rem = ( E_fail - E_i(t) ) / v_i(t)

[0060] Here, E_fail is a preset failure threshold. This invention relies solely on the long-term accumulated no-load reference value time series within the weighing device, executed locally on a microcontroller, to achieve low-cost predictive maintenance.

[0061] All fault detection, threshold calculation, self-healing algorithm, off-center load diagnosis, off-center load self-correction, mechanical structure early warning, and life prediction are completed internally by the device. This device uses a standard industrial communication interface and universal protocol. Users do not need to replace any existing sensors, junction boxes, or other hardware; they only need to replace their existing ordinary weighing device with this invention and add the corresponding status word reading program to the PLC or host computer. This will enable the entire weighing system to acquire complete intelligent diagnostic, self-healing, condition early warning, and predictive maintenance capabilities.

[0062] The beneficial effects of this invention include:

[0063] (1) By using a multi-cycle continuous judgment mechanism, 50 consecutive detection cycles (25 seconds) are used as the confirmation condition for permanent hardware faults, which effectively distinguishes between instantaneous electromagnetic interference and real faults, reducing the false alarm rate to near zero. At the same time, the fault window period is shortened from several hours to several days in the traditional scheme to 25 seconds.

[0064] (2) Through the adaptive reference value generation and dynamic maintenance algorithm, the no-load reference value is maintained independently for each sensor and the unidirectional drift and bidirectional fluctuation are distinguished. Only the continuous unidirectional drift such as material accumulation and aging is compensated, and the bidirectional fluctuation caused by temperature change is not responded to, thus eliminating the misjudgment and missed judgment caused by the fixed threshold.

[0065] (3) By using a short-term impact identification mechanism, temporary heavy-load impacts can be distinguished from permanent sensor damage, thus avoiding misjudgment of sensor failure due to instantaneous overload.

[0066] (4) By identifying jitter through multi-dimensional features, five-dimensional features such as amplitude, frequency, waveform factor, skewness and kurtosis are extracted to accurately distinguish jitter causes such as loose wiring, electromagnetic interference, mechanical vibration and sensor damage, and guide maintenance personnel to quickly locate faults.

[0067] (5) Through self-verification of data consistency of multiple sensors, sensor anomalies can be quickly identified in the absence of historical benchmark values, and environmental factors and sensor failures can be distinguished.

[0068] (6) By using the automatic differentiation and collaborative working logic of multiple factors, combined with unloading zero verification, multi-sensor synchronization verification, and dual verification of built-in temperature sensor, the sensor can automatically distinguish between permanent internal faults and external environmental factors such as material accumulation, temperature drift, and extreme weather.

[0069] (7) Through the data and logic isolation mechanism, independent storage areas are divided for each diagnostic module inside the microcontroller and staggered timing is adopted to ensure that multiple diagnostic algorithms run in parallel without interfering with each other.

[0070] (8) Through the off-center load self-diagnosis function, short-term fluctuations, material off-center loads and structural faults are distinguished according to the number of consecutive detections (less than 7 times, 7 to 11 times, 12 times and above), and the material off-center load is self-corrected online without the need for manual loading of weights.

[0071] (9) By automatically distinguishing between uneven force on the sensor and uneven material load, a diagnostic and early warning strategy is adopted for structural faults, and a combination of self-correction and prompting strategy is adopted for fluctuations in operating conditions, taking into account both production continuity and equipment safety.

[0072] (10) Through the hierarchical self-healing and dynamic reconstruction mechanism, temporary compensation and early warning are provided for short-term slight deviations, and abnormal channels are automatically removed for permanent deterioration or serious faults confirmed in the long-term trend. After the channel is restored to normal, it is automatically reintroduced into the weighing process, which not only ensures the continuity of production, but also ensures the reliability of measurement.

[0073] (11) By predicting the remaining lifespan of the sensor, and relying solely on the internally accumulated idle reference value time series, the microcontroller executes the operation locally, achieving low-cost predictive maintenance.

[0074] (12) All functions are integrated inside the device. It adopts standard industrial communication interface and general protocol. Users only need to replace the instrument body to enable the existing weighing system to obtain complete intelligent diagnosis, self-healing, early warning and predictive maintenance capabilities. It has the characteristics of low cost and easy deployment. Attached Figure Description

[0075] Figure 1 This is a two-dimensional schematic diagram of the overall layout of the device according to an embodiment of the present invention.

[0076] Figure 2 This is a block diagram of the overall system structure of the device according to an embodiment of the present invention.

[0077] Figure 3 This is a flowchart illustrating the power-on self-test and adaptive baseline value generation process in an embodiment of the present invention.

[0078] Figure 4 This is a flowchart illustrating the normal weighing and online synchronous monitoring process in an embodiment of the present invention.

[0079] Figure 5 This is a logic diagram for multi-cycle continuous judgment and interference filtering in an embodiment of the present invention.

[0080] Figure 6 This is a flowchart of the fault classification and self-healing process according to an embodiment of the present invention.

[0081] Figure 7 This is a flowchart illustrating the self-diagnosis process for off-center loading, misalignment, and mechanical structure abnormalities in an embodiment of the present invention.

[0082] Figure 8 This is a flowchart illustrating the sensor performance degradation early warning process in an embodiment of the present invention.

[0083] The markings in the attached diagram are explained as follows: 1—Weighing device power interface; 2—Weighing device button area; 3—Device communication interface; 4—Device display area; 5—Device mounting hole; 6—Device housing; 7—Weighing sensor interface-1; 8—Weighing sensor interface-2; 9—Weighing sensor interface-3; 10—Weighing sensor interface-4; 11—Weighing sensor cable-4; 12—Weighing sensor cable-3; 13—Weighing sensor cable-2; 14—Weighing sensor cable-1; 15—Weighing sensor-2; 16—Weighing sensor-3; 17—Scale body; 18—Weighing sensor-1; 19—Weighing sensor-4. Detailed Implementation

[0084] 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.

[0085] In the following embodiments, the sampling rate of the A / D conversion chip is preferably 512Hz, the resolution is preferably 24-bit, and the full-scale output code value is 2². 4 The microcontroller sequentially polls and reads the data from four sensors, updating the total weight data approximately every 8ms. Fault diagnosis uses an independent downsampling period, extracting one detection point every 0.5 seconds, and confirming the fault after 50 consecutive failures (25 seconds).

[0086] A preferred built-in 16M non-volatile memory is used for long-term storage of historical data, fault records, and configuration parameters of each channel sensor. The operating data of each functional module is stored in an independent address area of ​​the microcontroller's internal memory. Considering that the four sensors need to be stored independently, the address space is divided according to each channel with sufficient margin, as shown in the table below:

[0087] Functional modules Channel 1 address range Channel 2 address range Channel 3 address range Channel 4 address range illustrate Multi-period continuous judgment (sliding window) 0x1000~0x10FF 0x1100~0x11FF 0x1200~0x12FF 0x1300~0x13FF Each channel contains 256 bytes, storing S_i(t) and the number of exceptions. Adaptive baseline value maintenance (drift window M=5) 0x2000~0x20FF 0x2100~0x21FF 0x2200~0x22FF 0x2300~0x23FF Each channel contains 256 bytes, storing 5 sample values. Jitter detection (high-speed sampling window W=50) 0x3000~0x30FF 0x3100~0x31FF 0x3200~0x32FF 0x3300~0x33FF Each channel contains 256 bytes, storing 50 sampling points. Long-term jitter statistics (average amplitude buffer) 0x4000~0x40FF 0x4100~0x41FF 0x4200~0x42FF 0x4300~0x43FF Each channel contains 256 bytes, storing sliding window statistics. Off-center load diagnosis (sampling point buffer) 0x5000~0x50FF 0x5100~0x51FF 0x5200~0x52FF 0x5300~0x53FF 256 bytes per channel, written once every 10 seconds. Life expectancy prediction (daily historical data at midnight) 0x6000~0x6FFF 0x7000~0x7FFF 0x8000~0x8FFF 0x9000~0x9FFF Each channel contains 4096 bytes, enough to store approximately 10 years of daily data. Sensor uneven force detection (original snapshot) 0xA000~0xA0FF 0xA100~0xA1FF 0xA200~0xA2FF 0xA300~0xA3FF Each channel contains 256 bytes, storing a snapshot of μ_i(0). Multidimensional feature library (static data) 0xB000~0xBFFF (Shared area, storing feature library parameters) 4096 bytes, storing four fault type characteristics Event log area (impact, failure, environment) 0xC000~0xFFFF (Shared region, cyclic recording) 16384 bytes, recording various events and timestamps.

[0088] The aforementioned address allocation is based on the device's built-in non-volatile storage module. Data from each functional module is divided into contiguous address spaces within this module, with interconnected and non-overlapping address ranges. Daily midnight historical data and fault records for each channel can be permanently stored in this non-volatile storage module, meeting the needs of predictive maintenance throughout the device's lifecycle.

[0089] It should be noted that the multi-cycle continuous judgment mechanism, adaptive reference value generation and dynamic maintenance algorithm, hierarchical self-healing and dynamic reconstruction mechanism, off-center load self-diagnosis and self-correction method, sensor remaining life prediction method, multi-dimensional feature jitter identification method, and data and logic isolation mechanism described in this specification can all be independently implemented in the weighing device without relying on other diagnostic modules. Those skilled in the art can select one or more of these technical solutions to combine or implement individually according to actual needs.

[0090] The numerical parameters involved in the following embodiments include, but are not limited to, sampling rate 512Hz, detection period T_d=0.5 seconds, number of consecutive times L=50, learning rate η=0.8, drift detection window length M=5, jitter detection window length W=50, unidirectional drift judgment threshold 0.2%FS, loading and unloading confirmation threshold 0.1%FS, power-on deviation threshold 2%FS, single-channel deviation threshold 0.5%, impact strength threshold 30%, and adaptive jitter multiplier 1.5x (adjustable range is 1). The following parameters are preferred values ​​determined based on actual industrial field measurement data and standards such as IEC61000-4-4: 2 to 2.0 times and 4 times (adjustable range is 3 to 5 times), off-center load detection cycle T_b = 10 seconds, number of consecutive off-center load detections (less than 7 times for short-term off-center load, 7 to 11 times for moderate off-center load, and 12 times or more for severe off-center load), lifespan moving average days K = 7, lifespan warning threshold 0.05% / day, failure threshold 10%FS, moderate failure multiple 2 times, recovery cycle L_r = 100, etc. Those skilled in the art can adjust the above parameters within a reasonable range according to the electromagnetic environment, vibration level, accuracy requirements, and sensor characteristics of the actual application scenario. For example, the sampling rate can be selected within the range of 200Hz to 1000Hz, T_d can be selected within the range of 0.2 to 1.0 seconds, L can be selected within the range of 30 to 100, η ​​can be selected within the range of 0.3 to 1.0, and each threshold can fluctuate within a reasonable range according to the full scale and accuracy level. Such adjustments do not depart from the protection scope of this invention. The specific values ​​of each threshold can be determined based on the sensor model, range, accuracy level, and on-site environmental conditions of the actual application scenario. The fluctuation range is usually within ±50% of the preferred value.

[0091] Example 1: Online Diagnosis and Self-Healing of Sensor Faults in Smelting Enterprises

[0092] Take the newly built batching system of an industrial silicon smelting enterprise as an example. The plant has a 12-channel batching system, with raw materials packaged in various bins and 12 weighing devices centrally controlled by a PLC. The oil coke bin is supported by four load cells (distributed at the four corners), each with a rated capacity of 1 ton (full scale FS = 1000 kg). The original weighing devices collected the four signals through a junction box and then acquired them individually, only able to detect overall continuity.

[0093] After operating for a period of time, product quality continued to decline, but none of the equipment alarmed. The company invited technicians from various equipment manufacturers to conduct a joint diagnosis. Ultimately, by calibrating each device with weights, they discovered that after placing 80kg weights (20kg each at the four corners) on the oil coke bin, the device displayed only 57.5kg, a deviation of 22.5kg. Technicians spent half a day troubleshooting each component and confirmed that one sensor had no output at all (strain gauge detachment). Because this sensor failed, the 20kg load it originally bore was transferred to the other three sensors through the rigid structure of the scale body. However, due to mechanical friction and installation tilt, the transfer efficiency was not 100%, hence the measured value of 57.5kg was slightly lower than the theoretical value of 60kg. After replacing the sensor and recalibrating, the weighing returned to normal.

[0094] After replacing with the device of the present invention, see [link to relevant documentation]. Figure 1 The coke bin is supported by four weighing sensors (distributed at the four corners), each corresponding to... Figure 1 Weighing sensors -1 (18), -2 (15), -3 (16), and -4 (19) are included. The above four sensors are connected to the weighing sensor interfaces -1 (7), -2 (8), -3 (9), and -4 (10) on the outer shell (6) of the device, corresponding to channels 1 to 4. When the device is put into use for the first time, the empty scale is calibrated, and data for 32 sampling cycles are collected continuously. The average AD value of each sensor is taken as the initial empty reference value: μ_1(0)=0kg, μ_2(0)=0kg, μ_3(0)=0kg, μ_4(0)=0kg, and stored in the adaptive reference value maintenance area of ​​the corresponding channel in the non-volatile memory (channel 3 starting address 0x2200). The allowable deviation range is set to ±5% of the full scale, that is, [Δ_i⁻,Δ_i⁺]=[-50kg,50kg]. For detailed logic regarding power-on reference value restoration and loading / unloading confirmation, please refer to [link / reference]. Figure 3 .

[0095] The device employs a continuous 50-cycle judgment mechanism, with a detection cycle T_d = 0.5 seconds. During each detection, the average value within this 0.5-second period is extracted from the high-speed sampling data of the corresponding channel and used as the judgment input x_i(t). The overall process for normal weighing and online synchronous monitoring is described in [link to documentation]. Figure 4 For detailed logic on multi-period continuous judgment, please refer to [link / reference]. Figure 5When the sensor in channel 3 has no output due to cable corrosion, its weight value x_3(t) is always 0kg. Since the scale is empty at this time, μ_3(t) = 0kg, and the deviation x_3(t) - μ_3(t) = 0kg, which is still within the allowable range. However, the system detects that the weight value of channel 3 is always 0kg for 10 consecutive sampling periods (about 80ms), and the maximum change between adjacent sampling points is less than 0.01%FS (i.e., 0.1kg). Therefore, the system determines that the sensor is in an abnormal state of continuous no effective weight signal output, and the abnormal flag function f_3(t) is set to 1.

[0096] The sliding window counter S_3(t) begins to accumulate, and the calculation process is as follows: First detection cycle: f_3(1)=1, S_3(1)=1; Second detection cycle: f_3(2)=1, S_3(2)=2; and so on. Since S_3(t)=L=50, the cumulative abnormal duration reaches 25 seconds, the system determines that the channel 3 sensor has a permanent hardware failure, triggers the alarm "Channel 3 weighing sensor has a serious failure, please replace this weighing sensor" prompt, and enters the self-healing process. This judgment logic is described in [reference]. Figure 5 At this point, the device executes a moderate fault self-healing strategy: channel 3 data is completely removed, and the number of effective channels is adjusted from 4 to 3 when calculating the total weight. The total weight W_total = (x_1 + x_2 + x_4) / 3. For the complete fault-level self-healing process, please refer to [link to full details]. Figure 6 .

[0097] The fault window period has been shortened from several hours to several days in the traditional solution to approximately 25 seconds. During this period, products produced still pass inspection, avoiding significant economic losses. Simultaneously, weight data and status words are uploaded to the central control PLC in real time via serial communication / Ethernet. During subsequent production breaks, maintenance personnel replace the damaged sensor as prompted, and the engineer's workstation screen clearly displays "Channel 3 sensor fault, removed," achieving unattended remote monitoring.

[0098] During long-term operation, if the no-load value of a certain channel consistently exceeds the baseline value by 0.2%FS (i.e., 2kg) for five consecutive times, the baseline value will be adjusted upwards according to the adaptive formula μ_i←μ_i+0.8×(1 / 5)∑(x_{i,j}-μ_i), and a material accumulation warning will be displayed, making the measured value closer to the actual material weight. For the dynamic maintenance logic of the adaptive baseline value, please refer to [link / details]. Figure 3 For example, if the five consecutive no-load values ​​for channel 1 are 2.3kg, 2.5kg, 2.4kg, 2.6kg, and 2.5kg, respectively, all greater than μ_1+2kg, then the average drift is (2.3+2.5+2.4+2.6+2.5) / 5=2.46kg, the correction is 0.8×2.46≈1.97kg, and the new reference value is μ_1←0+1.97=1.97kg.

[0099] Example 2: Adaptive Baseline Value and Multi-level Safety Strategy for a 4-channel Sensor Batching System

[0100] Taking a 4-channel sensor dispensing system as an example, each sensor has a rated capacity of FS=100kg, and the allowable deviation range is set to ±5% (±5kg) of the full capacity. The original device used a fixed deviation threshold (±3%), which often resulted in a large difference in the weight values ​​of the sensors on both sides due to the frequent uneven loading of materials, leading to frequent false alarms. The device of this invention is now used.

[0101] Scenario 1: Adaptive baseline value to eliminate false alarms

[0102] After power-on self-test, the microcontroller loads the historical reference values ​​(all 0 kg) for each channel from the adaptive reference value maintenance area of ​​the corresponding channel in non-volatile memory, and measures the current no-load average value. If the deviation between the current no-load value x̄_i and the historical reference value δ_i > 2%FS (i.e., 2 kg) for a certain channel, the system waits for confirmation of the loading / unloading cycle. After confirmation, the reference value is updated. The power-on reference value recovery and verification process is detailed in [link to documentation]. Figure 3 Due to severe off-center loading on site, the system automatically widens the allowable deviation range to ±8% (user adjustable), corresponding to Δ_i⁻=-8kg and Δ_i⁺=+8kg.

[0103] During operation, a material imbalance caused the weight values ​​in channels 1 and 2 to be higher than expected, while those in channels 3 and 4 were lower than expected. Specifically, the values ​​were: x_1 = 15kg, x_2 = 14kg, x_3 = 5kg, and x_4 = 4kg. The deviations in each channel were within ±8kg, and the system did not trigger false alarms. The equipment operated continuously for several months without any false alarms. This scenario demonstrates that adaptive thresholds dynamically adjust according to on-site operating conditions, effectively eliminating false alarms caused by fixed thresholds.

[0104] Scenario 2: Early warning and graded handling of slight zero-point offset

[0105] After six months of operation, the sensor in channel 4 began to show zero-point drift due to the long-term dampness caused by water seepage from the material inside the silo. During 10 consecutive loading and unloading cycles with the scale empty, the system monitored in real time that the empty load values ​​of channels 1, 2, and 3 fluctuated within the range of 0 ± 0.1 kg, while the empty load value of channel 4 remained stable at around 0.6 kg each time.

[0106] Based on the multi-sensor consistency verification logic, the system determined that channel 4 exhibited a stable positive offset, but the offset was small (0.6%FS) and did not show a continuous increasing trend within 10 cycles. At this point, it was impossible to determine whether the offset was a permanent zero-point drift or a recoverable deviation caused by temporary moisture. The system issued a Level 1 warning: "Channel 4 has a slight zero-point offset; inspection recommended." However, the channel was not excluded from the weight calculation. Instead, the currently detected 0.6kg was used as a temporary compensation value, and this value was subtracted from the channel reading in subsequent weighings before being included in the total weight calculation. This approach ensured weighing accuracy while avoiding production interruptions due to minor anomalies.

[0107] The system continues to record the empty load value of channel 4 in each loading and unloading cycle, and archives the daily average empty load value to the remaining life prediction area (channel 4: 0x9000~0x9FFF). After about 5000 loading and unloading cycles (about 1 month), the system found through long-term trend analysis that the daily average empty load value of channel 4 has gradually increased from the initial 0.6kg to 1.2kg, with a drift rate of about 0.02kg / day, and shows a continuous unidirectional growth trend.

[0108] A moderate fault is identified when the drift exceeds twice the allowable range (0.5kg × 2 = 1.0kg). When the drift speed continuously exceeds the threshold and shows an accelerating trend, the sensor is deemed to have suffered irreversible performance degradation. The current drift of 1.2kg exceeds twice the allowable range, and the long-term trend confirms permanent zero-point drift. The system determines that the channel 4 sensor is severely degraded, issues a level 2 alarm ("Channel 4 sensor failed, please replace immediately"), and executes a moderate fault self-healing strategy: completely removes channel 4 data, adjusts the number of effective channels from 4 to 3 when calculating the total weight, and the total weight W_total = (x_1 + x_2 + x_3) / 3. See the fault classification self-healing process for details. Figure 6 .

[0109] This scenario demonstrates the invention's ability to differentiate between minor anomalies and severe faults: minor deviations in the short term are compensated temporarily to maintain production while continuously monitoring trends; once a long-term trend confirms permanent degradation, the system automatically escalates to a removal strategy and issues an alarm. This mechanism of "short-term fault tolerance + long-term trend judgment" avoids frequent downtime due to minor anomalies while ensuring timely response to genuine faults.

[0110] Scenario 3: Safety baseline during multiple severe failures

[0111] When the sensors in channels 1, 2, and 3 subsequently experience serious malfunctions (such as a constant weight value of zero), and the number of malfunctioning channels removed reaches the preset maximum value of 2, the device immediately stops its self-healing mode, issues an emergency shutdown alarm signal, and notifies the PLC to stop the batching process via the communication interface. This safety strategy is described in [reference needed]. Figure 6The system uses a cumulative elimination of channels as a branch for judgment. This scenario demonstrates the safety baseline of the self-healing mechanism: when the number of remaining normal channels is insufficient to guarantee metering reliability, the system refuses to continue self-healing and forces a shutdown for protection.

[0112] Example 3: Self-diagnosis of off-center loading, deviation, and mechanical structural anomalies, and self-correction of off-center loading.

[0113] A certain belt scale is supported by four load cells (distributed at the four corners), which respectively correspond to Figure 1 The load cells -1 (18), -2 (15), -3 (16), and -4 (19) are located on both sides of the belt (left channel 1, 2, right channel 3, 4), with a rated range FS=200kg. The system presets the off-center load detection cycle T_b=10 seconds, and calculates the mean x̄ and variance σ² of the weight values ​​of each channel every 10 seconds, as well as the maximum and minimum channel difference Δx_max-min. For the complete process of self-diagnosis of off-center load, misalignment, and mechanical structure abnormalities, please refer to [link to relevant documentation]. Figure 7 .

[0114] During normal operation, the weight values ​​of each channel are basically balanced. For example, in a certain sampling, the weight values ​​are: Channel 1: 80kg, Channel 2: 79kg, Channel 3: 81kg, and Channel 4: 80kg. The mean x̄ = (80 + 79 + 81 + 80) / 4 = 80kg, the variance σ² = [(0)² + (-1)² + (1)² + (0)²] / 4 = 0.5, and the maximum and minimum difference Δx_max - min = 81 - 79 = 2kg. Since both the variance and the difference are less than the preset threshold, the system determines that there is no off-center load.

[0115] When the material is biased to the left due to feeder wear, the weight values ​​of the two channels on the left are significantly higher than those on the right. In one sampling, the weights were: Channel 1: 102 kg, Channel 2: 100 kg, Channel 3: 64 kg, and Channel 4: 61 kg. The mean x̄ = (102 + 100 + 64 + 61) / 4 = 327 / 4 = 81.75 kg, and the variance σ² = [(20.25)² + (18.25)² + (-17.75)² + (-20.75)²] / 4 ≈ 370. The maximum and minimum difference Δx_max - min = 102 - 61 = 41 kg. Since the variance exceeds the preset off-center load threshold, the system determines it to be an off-center load condition.

[0116] Based on continuous detection results, the system performs graded processing of off-center load conditions:

[0117] Short-term off-center load: When the number of consecutive overloads in the same direction is less than 7, only the event is recorded and no alarm is triggered, so as to filter out short-term fluctuations.

[0118] Left and right off-center loading: The system independently maintains a 3-minute sliding window. When alternating left and right off-center loading is detected within the window, it is determined to be left and right off-center loading. At this time, the system outputs the prompt "Material off-center loading, please adjust the feeding device" and automatically starts the off-center loading self-correction function. The position weight coefficient β_i of each channel is obtained in advance by calibrating with standard weights at the four corners. Assuming β=[1.0,1.0,1.0,1.0], the dynamic compensation coefficient γ_i=(x̄ / x_i)·β_i is calculated. For channel 1: γ_1=(81.75 / 102)×1.0≈0.801, channel 2: γ_2=(81.75 / 100)×1.0=0.818, channel 3: γ_3=(81.75 / 64)×1.0≈1.277, channel 4: γ_4=(81.75 / 61)×1.0≈1.340. The corrected total weight of the scale is W_corrected=∑γ_i·x_i=0.801×102+0.818×100+1.277×64+1.340×61≈81.7+81.8+81.7+81.7=326.9kg, which is close to the actual total weight of 327kg.

[0119] Unilateral off-center load: When the number of consecutive out-of-range detections in the same direction reaches 7 or more, the unilateral off-center load level is judged. If the number reaches 7 but not 12, it is judged as moderate off-center load, and a warning of "the belt tension may be uneven, calibration is recommended" or "the belt is off-center, please check" is issued; if the number reaches 12 or more, it is judged as severe off-center load. Assuming the preset variance threshold is 100, the current variance is about 370, which exceeds twice the threshold (200) but does not reach four times the threshold (400), and an alarm of "the belt is seriously off-center, please adjust the belt tension" is issued. If the variance exceeds four times the threshold or the difference between the maximum and minimum channels exceeds 30% of the full scale, an emergency alarm of "the weighing frame may be deformed, stop the machine immediately for maintenance" is issued.

[0120] When the belt deviates to the right, the force on the right-side sensor increases, while that on the left decreases. In one sampling, the values ​​were: Channel 1: 67 kg, Channel 2: 66 kg, Channel 3: 93 kg, and Channel 4: 91 kg. The mean x̄ = (67 + 66 + 93 + 91) / 4 = 317 / 4 = 79.25 kg. The variance σ² increases significantly, and the load deviation is consistently heavier on the right side. The system outputs a warning or alarm of the appropriate level based on the number of consecutive detections and the degree of load deviation.

[0121] When uneven settlement occurs in the weighing frame foundation, the actual load borne by the left sensor decreases, while the load borne by the right sensor increases, but the total weight remains unchanged. The microcontroller, by analyzing the daily no-load zero-point historical records of each channel's remaining lifespan prediction area (Channel 1: 0x6000~0x6FFF, Channel 2: 0x7000~0x7FFF), found that the zero-point values ​​of Channels 1 and 2 continuously drifted negatively (decreasing), while the zero-point values ​​of Channels 3 and 4 continuously drifted positively (increasing), and this trend remained stable for three consecutive months. The load distribution was consistently uneven, with the left side being lighter and the right side heavier. Based on this, the system issued a warning: "Weighing frame structure abnormal, please check the foundation and welds." Maintenance personnel inspected and found that the left foundation had subsided, and promptly reinforced and repaired it, preventing further deformation of the weighing frame.

[0122] In scenarios where uneven sensor force is distinguished from material eccentricity:

[0123] Scenario 1 (Material Uneven Loading): The system detects that the uneven loading direction is not fixed, sometimes to the left and sometimes to the right, and the empty scale value is normal after unloading (zeroing deviation is less than 0.1%FS, i.e., less than 0.2kg). The system determines that the material is unevenly loaded, only outputs a prompt to adjust the feeding device, and does not trigger a fault alarm.

[0124] Scenario 2 (Uneven Sensor Force): The system detected that the unbalanced load direction was consistently heavier on the left side for 12 consecutive tests. After unloading, the empty scale showed that the empty load value of the left channel was higher: the empty load value of channel 1 was 0.4 kg, the empty load value of channel 2 was 0.3 kg, and the right channel was normal at 0.0 kg and 0.0 kg. The system read the original empty load reference value snapshot before compensation from the sensor force unevenness detection area (channel 1: 0xA000~0xA0FF, channel 2: 0xA100~0xA1FF): μ_1(0)=0 kg, μ_2(0)=0 kg. The mean value μ̄ = (0.4 + 0.3 + 0 + 0) / 4 = 0.175 kg is calculated. The variance of the empty scale σ_empty² = [(0.4 - 0.175)² + (0.3 - 0.175)² + (0 - 0.175)² + (0 - 0.175)²] / 4 = [0.0506 + 0.0156 + 0.0306 + 0.0306] / 4 ≈ 0.032. This exceeds the preset empty scale consistency threshold of 0.02, indicating deformation of the scale frame structure or uneven stress on the sensors. The deviation is calculated as d_1 = 0.4 / 200 × 100% = 0.20%, d_2 = 0.3 / 200 × 100% = 0.15%. Although the deviation of a single channel does not exceed 0.5%, the variance of the empty scale exceeds the limit, and the system issues a warning: "Uneven force on the left sensor, please check the mounting base." See [link to automatic differentiation logic] for details. Figure 7 Branch for determining the variance of the empty scale after unloading. Maintenance personnel found that the mounting base of the left sensor was loose. After tightening it, the empty load value returned to normal, and the off-center load alarm was cleared.

[0125] As can be seen from the above comparison, the system refuses to automatically compensate for uneven sensor stress and clearly prompts for mechanical maintenance, demonstrating a high degree of attention to structural safety. For material misalignment, it performs online self-correction to ensure continuous production, while simultaneously prompting the user to adjust the feeding device in a timely manner. This differentiated processing strategy balances production continuity and equipment safety, and is a concentrated manifestation of the intelligent diagnostic capabilities of this invention.

[0126] Example 4: Multidimensional Feature Jitter Recognition and Data Consistency Self-Verification

[0127] A channel 2 sensor in a batching system is experiencing intermittent signal jitter due to a loose wiring terminal. The microcontroller is collecting data at a 512Hz sampling rate. Because it sequentially polls four sensors, acquiring 50 consecutive sampling points per channel takes approximately 50 × 4 / 512 ≈ 0.39 seconds. Multidimensional feature parameters are calculated using these 50 sampling points as a window.

[0128] Example of data from 50 sampling points within the window (unit: kg, full scale FS=100kg): maximum value max=15.2, minimum value min=0.3, mean x̄=8.5, standard deviation σ=3.2.

[0129] Calculate the following characteristic parameters: jitter amplitude D_2 = max - min = 15.2 - 0.3 = 14.9 kg, relative to FS is 14.9%; jitter frequency f_2: the number of times the signal crosses the mean within the window N_zero = 8 times, the window duration T_w ≈ 0.39 seconds, f_2 = 8 / 0.39 ≈ 20.5 Hz; waveform factor W_f = X_rms / X_avg, where X_rms = √(∑x_i² / 50) ≈ 9.8, X_avg = 8.5, W_f ≈ 9.8 / 8.5 ≈ 1.15; skewness coefficient S_k = 1 / 50∑((x_i-8.5) / 3.2)³ ≈ 0.12; kurtosis coefficient K_u = 1 / 50∑((x_i-8.5) / 3.2) 4 -3≈0.48.

[0130] Compared with the preset feature library (stored at addresses 0xB000~0xBFFF), the highest matching degree was for the "loose wiring" type (loose wiring in the feature library corresponds to: low to medium amplitude 10%~20%FS, low frequency <50Hz, irregular waveform W_f≈1.0~1.3, kurtosis≈0). With consistent identification results across three consecutive windows, the system confirmed the jitter type as loose wiring and output a level one warning: "Channel 2 sensor signal is unstable, please check the wiring." Maintenance personnel checked according to the prompt and found the wiring terminals were loose; after tightening, the jitter disappeared.

[0131] Channel 4 of the same system experienced high-frequency spike jitter due to interference from a nearby frequency converter. The data within the window were: D_4≈8%FS, f_4≈120Hz, K_u≈4.2. A match with the feature library indicated "electromagnetic interference," and the device output a message: "Electromagnetic interference detected in Channel 4. Please check the shielding grounding." Maintenance personnel installed additional shielding, and the interference was eliminated.

[0132] In another scenario, after replacing a new sensor in a batching system, there are no historical baseline values ​​available. The device initiates a multi-sensor data consistency self-check. The empty scale values ​​(unit: kg) of the four sensors are: Channel 1: 0.5, Channel 2: 0.3, Channel 3: 0.4, and Channel 4: 0.6. The mean x̄ is calculated as (0.5 + 0.3 + 0.4 + 0.6) / 4 = 0.45, and the variance σ² is calculated as [(0.05)² + (-0.15)² + (-0.05)² + (0.15)²] / 4 = [0.0025 + 0.0225 + 0.0025 + 0.0225] / 4 = 0.05 / 4 = 0.0125. The preset consistency threshold T_consistency is 0.01. Since σ² > the threshold, a sensor malfunction is detected.

[0133] The following method was used to locate the problem: After removing channel 1, the remaining channel values ​​were 0.3, 0.4, and 0.6, with a mean of 0.433 and a variance of [(-0.133)² + (-0.033)² + (0.167)²] / 3 = [0.0177 + 0.0011 + 0.0279] / 3 ≈ 0.0156. The variance increased, so channel 1 was excluded. After removing channel 2, the remaining channel values ​​were 0.5, 0.4, and 0.6, with a mean of 0.5 and a variance of [0² + (-0.1)² + (0.1)²] / 3 = 0.02 / 3 ≈ 0.0067. The variance decreased significantly, indicating that the sensor in channel 2 was faulty. Upon inspection, it was found that the sensor in channel 2 was not properly installed; after reinstallation, it returned to normal.

[0134] Example 5: Short-term impact identification and automatic recovery of fault channels

[0135] During operation, a batching system experiences a sudden, heavy impact from material on sensor 2 in channel 2. The maximum weight value within a short time is x_2^peak = 145 kg, the full scale FS = 100 kg, and μ_2 = 0 kg. The impact intensity p_2 = (145 - 0) / 100 × 100% = 145% > 30%.

[0136] The system enters the impact confirmation process: waiting for the empty scale to complete the unloading process, and measuring the empty scale value x_2^after = 0.1kg. |x_2^after - μ_2| = 0.1kg, 0.1 / 100×100% = 0.1%FS < 0.2%FS, which is determined to be a temporary heavy load impact. No fault alarm is triggered, and only the event time and impact intensity are recorded in the event log area (address 0xC000~0xFFFF).

[0137] Several months later, the channel 3 sensor developed intermittent anomalies due to aging, with its anomaly flag f_3(t) intermittently set to 1. During multi-cycle continuous judgment, because the anomaly was discontinuous, S_3(t) remained less than 50, failing to trigger a permanent fault alarm. The handling logic for intermittent anomalies in multi-cycle continuous judgment is detailed below. Figure 5 However, the frequency of abnormalities increased significantly over a certain period, with 50 consecutive cycles showing abnormalities. This was deemed a permanent fault and the channel was removed. During self-healing mode operation, the microcontroller continued to monitor the status of channel 3 every 0.5 seconds. The fault status and recovery counter of this channel were temporarily stored in the multi-cycle continuous judgment buffer of the corresponding channel (channel 3: 0x1200~0x12FF). Within 100 consecutive detection cycles (approximately 50 seconds), f_3(t) was 0, and the sliding window count S_3(t) = 0, indicating that the channel had recovered. The system automatically reinstated channel 3 for weighing, restoring the denominator from 3 to 4. For the automatic recovery process of the faulty channel, please refer to [link to relevant documentation]. Figure 6 Meanwhile, the time of occurrence, duration (from removal to recovery), and type of the fault are stored in the event log area for future maintenance queries.

[0138] Example 6: Prediction of Remaining Sensor Lifespan

[0139] Taking a certain batching system as an example, a sensor that has been used for two and a half years has a rated range FS=500kg. The microcontroller temporarily stores the zero-point data of the most recent K days in the remaining life prediction area of ​​the corresponding channel (e.g., channel 1: 0x6000~0x6FFF) for moving average calculation. The daily zero-point value is archived to this area for long-term storage after moving average processing. Each channel independently stores one historical record. The initial no-load reference value μ_i(0)=0kg. For the complete process of sensor remaining life prediction, please refer to [link to relevant documentation]. Figure 8 .

[0140] Under normal conditions, the zero-point drift rate of this sensor is approximately 0.01% / day, or 0.05 kg per day. Starting from the 28th month (approximately day 840), the microcontroller detected a gradual increase in the drift rate of this channel, and the recorded data are as follows: Day 840 to Day 846: average daily drift 0.10 kg (0.02% / day); Day 847 to Day 853: average daily drift 0.20 kg (0.04% / day); Day 854 to Day 860: average daily drift 0.30 kg (0.06% / day).

[0141] The average drift velocity v_i(t) over the most recent K=7 days was calculated using the moving average method. On day 860, v_i = 0.06% / day, and it has exceeded the preset warning threshold of 0.05% / day for 7 consecutive days. The logic for this judgment is described in [link to relevant documentation]. Figure 8 The system detects that the sensor's performance has entered a period of accelerated degradation and issues a warning: "Sensor performance in channel X has deteriorated; replacement within the estimated number of days is recommended."

[0142] Estimated remaining usable days: Preset failure threshold E_fail = 10%FS = 50kg, current E_i(t) = μ_i(t) / FS × 100%. Assuming current μ_i(t) = 8kg, E_i(t) = 8 / 500 × 100% = 1.6%, then L_rem = (10% - 1.6%) / 0.06% = 8.4 / 0.06 = 140 days. See [link to calculation process] for details. Figure 8 The remaining lifespan estimation step is as follows. The device interface displays the performance trend curve and the estimated remaining days.

[0143] After receiving the warning, maintenance personnel replaced the sensor during the end-of-month shutdown for maintenance, thus preventing a production stoppage due to a sudden malfunction during peak production periods. After replacement, the zero-point drift rate of the new sensor returned to normal (0.01% / day), and the warning was cleared.

[0144] Example 7: Automatically distinguishing between material accumulation and sensor zero-point drift

[0145] A batching system has been running continuously for several weeks. The system monitors the AD values ​​of each channel and the changing trends of the total empty weight of the entire scale in real time. On a certain day, the system detected in real time that the total empty weight of the entire scale after unloading was consistently higher than the historical benchmark value, with a current offset of 0.3 kg (FS=200 kg, 0.15% higher than FS). The system immediately initiated a multi-factor analysis process, retrieving AD value data for each channel in real time and continuously analyzing the contribution distribution of each channel to the total offset.

[0146] Scenario 1 (Uniform Material Accumulation): Real-time system analysis revealed that although the total weight increased by 0.3kg, this 0.3kg increment was relatively evenly distributed across the channels. The system calculated the empty offset contribution of each channel in real time as follows: Channel 1 approximately +0.08kg, Channel 2 approximately +0.07kg, Channel 3 approximately +0.08kg, and Channel 4 approximately +0.07kg, totaling 0.30kg, consistent with the actual total offset. The system continuously calculated the variance of the offset for each channel: mean μ̄ = 0.30 / 4 = 0.075kg, variance σ_empty² = [(0.005)² + (-0.005)² + (0.005)² + (-0.005)²] / 4 ≈ 0.000025, indicating extremely small variance. Simultaneously, the system calculated the correlation coefficient of the offset for each channel in real time; the correlation coefficients between channels were all close to 1 (showing a synchronous changing trend). Based on this, the system determines in real time that the material is accumulating evenly and issues a prompt that "there may be material accumulation on the scale body, and it is recommended to clean it."

[0147] Subsequently, the system continuously monitors whether the total unloaded weight meets the compensation conditions. In subsequent loading and unloading cycles, the system detects that the total unloaded weight deviation after five consecutive unloading cycles exceeds 0.2%FS (0.4kg), specifically 0.42kg, 0.45kg, 0.48kg, 0.44kg, and 0.46kg. The system calculates the average of these five deviations in real time as (0.42+0.45+0.48+0.44+0.46) / 5=0.45kg, with a learning rate η=0.8 and a correction amount of 0.8×0.45=0.36kg. The system then updates the baseline value μ_i←μ_i+0.36kg in real time (see the automatic zero-point drift compensation process). Figure 3 (Medium- to long-term operation compensation branch). After the user cleans the material from the scale during the downtime as prompted, the system detects in real time that the A / D values ​​of each sensor on the empty scale have returned to the normal range, immediately cancels the correction value, and automatically clears the prompt.

[0148] Scenario 2 (Single Sensor Moisture): Real-time system analysis revealed that the total offset of 0.3 kg was almost entirely contributed by channel 3. The system continuously tracked the no-load offset of channel 3 and calculated its drift velocity in real time. Currently, the offset of channel 3 is approximately +0.3 kg, while the contributions from other channels are close to 0 kg. The system calculated the average drift velocity of the group, v̄(t), to be approximately 0.01% / day, and the drift velocity of channel 3, v_3(t), to be 0.08% / day. The deviation Δv_3 = |0.08 - 0.01| = 0.07, exceeding twice the standard deviation of 0.04. The system immediately detected an anomaly in channel 3 and issued an alarm: "Channel 3 sensor zero point anomaly, please check." Upon inspection, it was found that the weighing sensor -3 was damp; after replacement, normal operation was restored.

[0149] Scenario 3 (Uniform Temperature Drift): During a sudden drop in winter temperature, the system monitors in real time that the total weight of the empty scale decreases synchronously, with a current decrease of 0.2 kg (0.1%FS). The system analyzes the contribution distribution of each channel in real time and finds that the offset distribution of each channel is uniform, with a variance σ_empty²≈0 and a correlation coefficient close to 1. At the same time, the system automatically retrieves data from the built-in temperature sensor and finds that the ambient temperature is decreasing synchronously, and the drift direction of the empty values ​​of each channel is consistent with the direction of temperature change (cooling → negative drift). By calculating the correlation coefficient between the drift speed and the rate of temperature change, r=0.88 is obtained, which exceeds the preset threshold of 0.8. Based on this, the system confirms it as temperature drift and does not trigger a fault alarm. Since the drift amplitude of 0.2 kg is less than the zero-point drift compensation activation threshold of 0.2%FS (0.4 kg), the system only records the event in the event log area and does not activate compensation. After the temperature rises, the system monitors in real time that the total weight of the empty scale automatically recovers to the original baseline value without any manual intervention.

[0150] Example 8: Automatic differentiation between extreme weather and material accumulation

[0151] A factory in northern China experienced a cold wave during winter, with the ambient temperature plummeting within 24 hours. An internal temperature sensor recorded the temperature change curve and stored it in the event log area (addresses 0xC000 to 0xFFFF).

[0152] The system monitors the changing trend of the empty zero point of each channel in real time. During the cold wave, the system detected that the empty zero points of channels 1-4 decreased synchronously and continuously, with a real-time drift speed of approximately 0.1% / hour, and a daily decrease of approximately 0.3%FS (FS=200kg, a decrease of 0.6kg / day). Multi-sensor consistency verification showed no significant change in variance σ_empty², and the correlation coefficient of each channel was >0.95. The system determined in real time that this was due to the influence of ambient temperature and did not issue a material accumulation alarm, only... Figure 1 The device displays a message in area (4) stating "The ambient temperature changes significantly, which may affect the weighing data." Since temperature drift is a bidirectional fluctuation and does not meet the compensation condition of "continuous unidirectional" drift, the system does not initiate automatic zero-point drift compensation.

[0153] After the cold wave subsided, continuous monitoring by the system revealed that the zero points of channels 1-4 remained synchronously high under no-load conditions and did not return to zero after unloading. At this point, the real-time calculated drift speed had decreased to approximately 0.01% / hour. The system detected that this low-speed drift had persisted for over 24 hours, and determined in real-time that the drift nature had changed from bidirectional fluctuations caused by temperature to continuous unidirectional drift caused by material accumulation. The system issued a "Material accumulation on the weighing scale may exist; cleaning is recommended" prompt and initiated automatic zero-point drift compensation to gradually correct the issue. After the user cleaned the accumulated material, the no-load zero point returned to normal.

[0154] This embodiment demonstrates the ability of the present invention to automatically distinguish between temperature drift and material accumulation under extreme weather conditions: it only provides a warning without compensation for rapid bidirectional fluctuations caused by temperature, while actively compensating for continuous unidirectional drift caused by material accumulation, reflecting the design concept of selective compensation.

[0155] Example 9: Built-in temperature sensor assists in identifying temperature drift

[0156] A factory in southern China experienced a sudden rise in ambient temperature during the summer heat. The built-in temperature sensor (installed inside the housing near the stainless steel sidewall) recorded the continuous temperature increase, and the data was stored in the event log area (addresses 0xC000 to 0xFFFF).

[0157] Meanwhile, multi-sensor consistency verification shows that all channels have a zero-point synchronous positive drift of approximately 0.15%FS (FS=200kg, drift 0.3kg) under no-load conditions. The system automatically retrieves temperature sensor data and continuously compares it: the drift direction is consistent with the temperature change direction (heating → positive drift); the system maintains a 24-hour sliding window in real time, recording a data point of drift rate versus temperature change rate every hour, and continuously calculating the correlation coefficient within the window. When the correlation coefficient r exceeds the preset threshold of 0.8, it is confirmed as temperature drift. The current window calculates r=0.85, exceeding the threshold, but the system does not trigger a fault alarm, only recording "significant changes in ambient temperature, weighing data may be affected" in the log. Since temperature drift is a bidirectional fluctuation, it does not meet the "continuous unidirectional" compensation condition, and the system does not initiate automatic zero-point drift compensation. After the temperature drops, the zero point automatically recovers without any manual intervention.

[0158] Taking the batching system of an industrial silicon smelting enterprise in Example 1 as an example, the traditional solution, where the device still displays "normal" even after the sensor completely fails, means the factory can only discover the fault by checking the weights or tracing back after a product problem occurs, resulting in economic losses. With the device of this invention, sensor faults can be automatically identified and alarmed within approximately 25 seconds. If downtime is not permitted on-site, a self-healing mode can be automatically activated to maintain weighing continuity. Similar scenarios are common in industries such as cement, chemicals, feed, and metallurgy. This device can be used for low-cost intelligent transformation of old systems and for deployment in new systems, upgrading the traditional "post-event tracing" to "online real-time monitoring" and "proactive maintenance based on equipment health status," demonstrating significant industrial practical value and broad application prospects.

[0159] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent weighing device based on multi-cycle continuous judgment, comprising a device housing, multiple sensor input interfaces disposed on the housing, M independent filtering and amplification circuits disposed within the housing, an M-channel A / D conversion chip, a microcontroller, a display and alarm unit, a data storage unit, and a communication interface; characterized in that: The microcontroller is configured to: extract feature values ​​from each sensor signal at a preset detection cycle, compare the feature values ​​with the current no-load reference value, and mark an anomaly when the deviation exceeds the allowable deviation range. Maintain a sliding window of preset length and count the number of consecutive anomalies; When the number of consecutive abnormalities reaches a preset number, it is determined that the sensor has a permanent hardware failure, triggering the self-healing process and outputting an alarm signal through the display alarm unit; When the number of consecutive abnormalities is less than the preset number, it is determined to be a momentary disturbance, no alarm is triggered, and the counter is automatically cleared in the subsequent normal cycle; The self-healing process includes performing one of the following operations based on the severity of the fault: attenuating and weighting the faulty channel data, completely removing the faulty channel data, or performing weighted compensation based on the remaining normal channels.

2. The intelligent weighing device according to claim 1, characterized in that: The preset detection cycle is 0.5 seconds, the preset length is 50, the preset number of times is 50, and the total fault confirmation time is 25 seconds.

3. The intelligent weighing device according to claim 1, characterized in that: The microcontroller is also configured to execute an adaptive reference value generation and dynamic maintenance algorithm, including: measuring the empty load value of each channel in the empty scale state after each loading and unloading cycle is completed; determining the offset direction and persistence of the current empty load value relative to the empty load reference value; when it is detected that the empty load value exceeds the preset offset threshold in the same direction for a preset number of consecutive preset number of times, it is determined to be a continuous unidirectional drift, and the reference value is corrected with a preset learning rate; when it is detected that the empty load value fluctuates in both directions, the reference value is kept unchanged.

4. The intelligent weighing device according to claim 3, characterized in that: The number of consecutive preset times is 5, the preset learning rate ranges from 0 to 1, and the preset offset threshold is 0.2%FS.

5. The intelligent weighing device according to claim 1, characterized in that: The self-healing process specifically includes: for minor deviations that occur in the short term, the currently detected deviation is used as a temporary compensation value to correct the channel reading, a first-level warning is issued, and its long-term trend is continuously monitored; when the long-term trend analysis confirms that the deviation has continued to increase and exceeds a preset multiple threshold, the channel data is completely removed and a second-level alarm is issued; when the number of removed faulty channels reaches a preset upper limit, an emergency shutdown alarm is issued.

6. The intelligent weighing device according to claim 5, characterized in that: The preset multiple threshold is twice the allowable deviation range; when no abnormality is detected in the removed channel within a continuous preset recovery period, the channel is automatically reinstated for weighing, wherein the preset recovery period is 100 detection cycles.

7. The intelligent weighing device according to claim 1, characterized in that: The microcontroller's internal memory is divided into multiple independent address regions, including a sliding window buffer for the multi-cycle continuous judgment module, a drift detection window buffer for the adaptive reference value maintenance module, a sampling point buffer for the off-center load diagnosis module, and a snapshot area of ​​the original no-load reference value for the sensor force unevenness detection module. The threshold parameters of each module are stored in different sectors of non-volatile memory and do not overwrite each other. At the timing level, each functional module adopts a staggered processing mechanism: A / D sampling is performed according to the first preset cycle, multi-cycle continuous judgment is performed according to the second preset cycle, off-center load diagnosis is performed according to the third preset cycle, and adaptive baseline value maintenance is performed when the unloading event is triggered.

8. The intelligent weighing device according to claim 1, characterized in that: The microcontroller is also configured to perform an off-center load self-diagnosis function, which calculates the mean and variance of the weight values ​​of each channel according to a preset off-center load detection cycle. Based on the continuous detection results, if the first detection exceeds the limit but recovers on the second, only the event is recorded. If the third or more detections exceed the limit and the direction is not fixed, it is determined that the material is off-center and online self-correction is initiated. If the fourth or more detections exceed the limit and the direction is fixed, it is determined that the structure is faulty and a graded alarm is output.

9. The intelligent weighing device according to claim 8, characterized in that: The preset off-center load detection cycle is 10 seconds, the first count is 6, the second count is 7, the short-term off-center load corresponds to less than 7 consecutive over-limit times, the moderate off-center load corresponds to 7 to 11 consecutive over-limit times, and the severe off-center load corresponds to 12 or more consecutive over-limit times; the online self-calibration corrects each channel through a dynamic compensation coefficient γ_i=(x̄ / x_i)·β_i, where x̄ is the average weight of each channel and β_i is the pre-calibrated position weight coefficient of each channel.

10. An online fault diagnosis and self-healing method for an intelligent weighing device based on multi-cycle continuous judgment, applied to the intelligent weighing device according to any one of claims 1 to 9, characterized in that, Includes the following steps: Step 1: Extract feature values ​​from each sensor signal at a preset detection cycle, compare the feature values ​​with the current no-load reference value, and mark an anomaly when the deviation exceeds the allowable deviation range; Step 2: Maintain a sliding window of a preset length and count the number of consecutive anomalies; Step 3: When the number of consecutive abnormalities reaches the preset number, it is determined that the sensor has a permanent hardware failure, triggering the self-healing process and outputting an alarm signal; When the number of consecutive abnormalities is less than the preset number, it is determined to be a momentary disturbance, no alarm is triggered, and the counter is automatically cleared in the subsequent normal cycle; Step 4: The self-healing process includes performing one of the following operations based on the severity of the fault: attenuating and weighting the faulty channel data, completely removing the faulty channel data, or performing weighted compensation based on the remaining normal channels.

Images