Dynamic load monitoring lithium battery lifting appliance management system

Abstract

The invention belongs to the technical field of industrial lifting equipment monitoring, and particularly relates to a dynamic load monitoring lithium electric lifting appliance management system which comprises a data acquisition module, a signal processing module, a cloud processing module, a load early warning module, an execution control module and a data storage module. The data acquisition module adopts a framework formed by an acquisition end and a cloud end; the signal processing module adopts an anti-aliasing filter and a wavelet packet to decompose and process load data; the cloud processing module dynamically adjusts computing resources by adopting a distributed architecture and performs load prediction by utilizing a pre-training model; the load early warning module performs normalization processing on the data and constructs a dynamic confidence interval; the execution control module awakens the power module according to the early warning signal and dynamically adjusts the weight of the abnormal item; the data storage module stores historical fault data to optimize early warning accuracy. According to the system, high-precision real-time monitoring of the load of the lithium battery lifting appliance can be realized, and the safety of lifting operation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a sling management system, in particular to a dynamic load monitoring lithium battery sling management system. BACKGROUND

[0002] With the continuous improvement of industrial automation and intelligence, lithium battery sling as an important part of modern industrial handling equipment, its load monitoring and management system performance is directly related to production safety and efficiency. At present, the lithium battery sling load monitoring system mainly realizes the monitoring and early warning of the sling running state by collecting and analyzing the load data.

[0003] The existing load monitoring system mostly adopts the architecture composed of the acquisition end and the cloud end, and transmits the load data to the cloud end for processing and analysis through the network. For example, CN120498121A discloses an intelligent control and management system for power supply based on dynamic load identification, which includes a data acquisition module, a signal processing module, a dynamic load identification module, etc., for high-frequency sampling and processing of voltage, current, power factor and other data in the power supply line. The online evaluation system for the running state of the electric energy metering device disclosed in CN120031545A adopts a dynamic load balancing mode for electric energy metering data packet transmission, which improves the response speed and processing capacity of the system.

[0004] In terms of load balancing control, CN119341180A proposes an adaptive load balancing ring network cabinet, which intelligently optimizes load distribution through real-time data and edge computing, integrates PID control and fuzzy logic, and dynamically adjusts parameters to ensure efficient operation of the power system. These technologies provide important references for the development of load monitoring systems.

[0005] In terms of data processing and anomaly detection, CN120358058A discloses a power stealing prevention electric energy information acquisition system based on an outlier algorithm model, which analyzes user power consumption information through an outlier algorithm and can identify abnormal behavior. CN120750982A proposes an electric energy meter remote monitoring and management system based on edge computing, which innovatively adopts a sliding window adaptive threshold and edge lightweight SVM linkage algorithm to dynamically calculate data features and adaptively adjust the threshold to preliminarily screen out anomalies.

[0006] However, the existing lithium battery sling load monitoring system still has the following technical problems:

[0007] Firstly, the existing system mostly adopts the architecture composed of the acquisition end and the cloud end, and transmits the load data to the cloud end for processing through network transmission, which has a one-way delay of up to 100-200 ms, far exceeding the 50 ms response upper limit required for sling dynamic monitoring, resulting in a lag of the early warning signal behind the actual load change. This delay may cause safety hazards during critical operations such as lifting heavy objects with the sling.

[0008] Secondly, the traditional load monitoring and management method mainly relies on static threshold triggering switching or a single parameter as the basis for judgment, which cannot fully reflect the real health status of the spreader system. This simple judgment mechanism is easy to produce misjudgment or omission under complex working conditions.

[0009] Thirdly, in terms of sensor accuracy, the wavelength resolution of ordinary spectrum demodulator is only 5-10pm, which cannot capture the small wavelength drift corresponding to the 0.1%FS precision of FBG sensor, and when the sampling rate is lower than 500Hz, it cannot restore the load impact peak value in the instant of lifting the spreader. This leads to the system cannot accurately capture the small changes and transient characteristics of the load.

[0010] Fourthly, in the construction process of the multi-dimensional data fusion monitoring system, the sensor layout and data acquisition method still need to be further optimized to improve the comprehensiveness and accuracy of the monitoring system. Single data source is difficult to cope with complex and variable working conditions.

[0011] Finally, in terms of algorithm optimization, the accuracy and real-time performance of load identification still need to be improved, which is difficult to meet the complex and variable load monitoring requirements. Especially in the case of environmental mutation or load anomaly, the lack of adaptive adjustment mechanism affects the reliability of the system.

[0012] Therefore, it is urgent to develop a dynamic load monitoring lithium battery spreader management system that can realize high precision, low delay and multi-dimensional data fusion, in order to improve the accuracy and real-time performance of load monitoring and ensure the safe and reliable operation of the spreader. SUMMARY

[0013] The purpose of the present application is to solve the technical problems raised in the background art by setting up a dynamic load monitoring lithium battery spreader management system.

[0014] To solve the above technical problems, the technical scheme is as follows: a dynamic load monitoring lithium battery spreader management system, comprising a data acquisition module, a signal processing module, a cloud processing module, a load warning module, an execution control module and a data storage module.

[0015] The data processing module is integrated in a control unit of a cloud processing module, and is connected in real time with a data acquisition module (communication delay is not greater than 10 ms) and a signal processing module (data throughput is not less than 1 Mbps) through an Ethernet interface, and a built-in logic controller (such as an STM32H7) in the data processing module executes a sequential execution process of data receiving, checking and distribution, and specifically includes: receiving clean data of the signal processing module and forwarding the clean data to a cloud computing node, synchronously calling historical error data from a data storage module and pushing the historical error data to a load warning module, and realizing time sequence alignment (synchronous error is not greater than 5 ms) of a full-process data link.

[0016] The data acquisition module adopts a 1 kHz high sampling rate ADC to acquire load data, and transmits the load data to the cloud processing module through a network; a high-speed differential probe with a sampling rate not less than 1 MHz is used to capture the load signal in a full cycle, so that the micro changes of the load can be captured.

[0017] Further, the signal processing module performs filtering processing on the acquired load data, uses an anti-aliasing filter with a cutoff frequency of 40% of the sampling rate to effectively suppress high-frequency noise interference, and simultaneously uses wavelet packet decomposition to strip the load waveform of high-frequency noise, effectively suppressing electromagnetic interference pollution to the detection system.

[0018] Preferably, the cloud processing module adopts a distributed architecture, dynamically adjusts the computing resources of each node according to the real-time nature of the power grid and the real-time changes of the load, uses pre-trained general image recognition model parameters, and only performs special training on the full connection layer for the load prediction task, thereby significantly reducing the demand for labeled data.

[0019] Optionally, the load warning module performs normalization processing on the real-time acquired load data, and unifies the dimension to the range of 0 to 100%; based on the historical prediction error range, a curve upper and lower boundary is generated to form a dynamic confidence interval, and an adaptive prediction fault tolerance space is constructed, so that when the environment mutates or the load is abnormal, the system can automatically reduce the certainty weight of the prediction result.

[0020] Further, the execution control module wakes up the dormant power module in time according to the warning signal output by the load warning module; if the load suddenly increases or the ripple is abnormal, the weight of the abnormal item is increased to 50%, and the weights of the remaining indicators are compressed in proportion, so as to enhance the sensitivity to hardware faults.

[0021] Preferably, the system further comprises a data storage module for storing voltage waveforms, load responses and environmental data in a historical fault library, and through analysis and mining of these data, the accuracy and real-time performance of the load warning are further optimized.

[0022] Advantages of the present application:

[0023] By adopting the load data under the high sampling rate of 1kHz to transmit to the cloud for processing, the one-way delay is significantly reduced, the early warning signal can timely reflect the actual load change, and the problem that the early warning signal lags behind the actual load change in the prior art is effectively solved; by improving the sampling rate and wavelength resolution, in combination with the high-precision measurement capability of the FBG sensor, the accurate capture of the slight wavelength drift is realized, the wavelength resolution deficiency of the traditional spectrum demodulator is overcome, and the accuracy of the load monitoring is improved; the load impact peak value at the moment of lifting the lifting appliance can be restored, the information loss problem caused by the low sampling rate in the prior art is effectively solved, and the integrity and reliability of the load monitoring are improved; by optimizing the sensor layout and data acquisition mode, the comprehensive fusion monitoring of multidimensional data is realized, the limitation of single parameter judgment in the prior art is overcome, and the real health status of the power supply system is more comprehensively reflected; according to the real-time of the power grid and the real-time change of the load, the cloud computing resources are dynamically adjusted, the optimal allocation of resources is realized, and the overall operation efficiency and response speed of the system are improved. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0025] In the drawings:

[0026] Figure 1 is a flowchart of the present application;

[0027] Figure 2 is a system hardware module relationship flowchart in the present application;

[0028] Figure 3 is a control method step flowchart in the present application. DETAILED DESCRIPTION

[0029] The exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, which include various details of the embodiments of the present disclosure to help understanding, and should be considered as merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, for the sake of clarity and conciseness, the description below omits the description of well-known functions and structures.

[0030] The specific embodiments are given below.

[0031] Embodiment one

[0032] The application provides a dynamic load monitoring lithium battery sling management system, which comprises a data acquisition module, a signal processing module, a cloud processing module, a load warning module, an execution control module and a data storage module.

[0033] The signal flow of each module is as follows: data acquisition module; signal processing module (filter fusion); data processing module (integration and distribution); cloud processing module (prediction) / data storage module (archiving); load warning module (judgment); execution control module (response), forming a closed-loop control link, and the total delay is not more than 200 ms.

[0034] The data acquisition module adopts an architecture composed of an acquisition end and a cloud end, configures an ADC with a sampling rate of 1 kHz for collecting sling load data, and transmits the load data to the cloud processing module through a network. The ADC sampling accuracy is 12 bits, which can accurately capture the load change signal. The data acquisition module also includes a high-speed differential probe with a sampling rate of 1 MHz, which is used to capture the load signal in the whole cycle to capture the slight changes in the load. The high-speed differential probe and the ADC work together to realize synchronous acquisition of the load transient change signal, ensuring that the system can comprehensively monitor the load state of the sling under various working conditions.

[0035] The data acquisition module also associates the wavelength drift data of the FBG sensor, uses a quartz-based FBG sensor, the wavelength resolution is not more than 1 pm (satisfying the 0.1% FS accuracy requirement), the sampling rate is not less than 1 kHz (matching the ADC sampling rate), and is installed in the strain concentration area of the sling load beam (such as the 1 / 4 position of the beam end). The thermal expansion coefficient of the quartz-based FBG is:

[0036] α=5.5×10 -7 / ℃;

[0037] Thermal light coefficient:

[0038] ξ=6.6×10 -6 / ℃;

[0039] The GPS time module is used to realize the millisecond-level timestamp alignment of the FBG and the ADC, and the synchronization error is not more than 1 ms. The data frame format adopts the structure of timestamp + wavelength value + load value. The data frame format is defined as: timestamp (13 bits) + wavelength value (16 bits) + load value (16 bits). In order to ensure the long-term stable work of the FBG sensor in the industrial hoisting vibration, oil stain and temperature sudden change environment, the installation and protection scheme is as follows:

[0040] Fixing process and material parameters: A three-stage fixing scheme is adopted: sandblasting pretreatment + structural adhesive bonding + mechanical reinforcement. The mounting surface of the load-bearing beam is first sandblasted to a roughness of Ra1.6μm to remove oxide layer and oil stains; epoxy structural adhesive (model: 3MDP460) is applied to the bottom surface of the sensor (adhesive thickness 0.2mm), and after bonding, a pressure of 0.5MPa is applied and maintained for 24 hours to ensure that the shear strength between the sensor and the beam is not less than 20MPa; the sensor lead wire end is further fixed to the beam with M2 stainless steel screws (pre-tightening torque 5N·m) to prevent vibration from causing sensor displacement.

[0041] Environmental protection structure: Customized 304 stainless steel protective shell (size: 30mm×15mm×8mm, thickness 1mm), the shell is filled with a silicone buffer pad with a hardness of 50 Shore A (covering the entire sensor area, thickness 2mm) to buffer vibration and impact; EVA sponge pad (thickness 5mm) is added at the contact point between the protective shell and the beam to reduce the transmission of beam vibration; the sensor lead wire adopts oil-resistant shielded cable (model: PUR sheathed shielded wire, wire diameter 0.5mm²), and the cable joint is sealed with IP67 waterproof sealant (model: Loctite 5699) to prevent oil and moisture intrusion.

[0042] Temperature compensation calibration process: After system installation, pre-calibration is performed at three reference temperature points: -10℃, 25℃, and 50℃. The lifting device is placed in a constant temperature chamber, and after the temperature stabilizes, the FBG wavelength value is collected and calculated using the formula:

[0043] Δλ calibration = 0.02 × ΔT (unit: pm / ℃, ΔT is the difference between the actual temperature and 25℃);

[0044] A temperature compensation table is generated and pre-written into the Flash memory of the signal processing module to correct temperature drift errors in real time.

[0045] The signal processing module fuses the wavelength drift data with the load data acquired by the ADC, using the formula:

[0046] ΔλB / λB=(1-pe)ε+(α+ξ)ΔT;

[0047] The strain and temperature effects are separated (where pe is the photoelastic coefficient of 0.22, ε is the strain, α is the coefficient of thermal expansion, ξ is the thermo-optic coefficient, and ΔT is the temperature change) to correct for load measurement errors. Network transmission uses 4G or industrial Ethernet with a transmission rate of no less than 1Mbps, and is configured with a data encryption protocol (AES-256) to ensure data security.

[0048] The signal processing module filters the load data output from the data acquisition module, including using an anti-aliasing filter to suppress high-frequency noise interference. The cutoff frequency of this anti-aliasing filter is set to 40% of the ADC sampling rate, i.e., 400Hz. The anti-aliasing filter uses a 4th-order Butterworth low-pass filter (passband flatness adapted to load signal characteristics), with a cutoff frequency of 400Hz (corresponding to an ADC sampling rate of 1kHz), a stopband attenuation of not less than -60dB, and suppression of high-frequency interference above 1kHz. When the grid frequency fluctuation is no greater than ±0.5Hz, wavelet packet decomposition (scale parameter 6) is enabled; when the fluctuation is greater than ±0.5Hz, it automatically switches to a digital notch filter (adaptively tracking frequency drift). Simultaneously, wavelet packet decomposition is used to strip high-frequency noise from the load waveform to suppress electromagnetic interference. The scale parameter of the wavelet packet decomposition is 6 to extract the transient characteristics of the load signal. The signal processing module also includes a digital notch filter, which uses an adaptive algorithm to track power grid frequency fluctuations in real time to dynamically adjust the notch center frequency, eliminating the 50Hz sinusoidal fundamental component. The process involves anti-aliasing filtering → digital notch filtering / wavelet packet decomposition → signal normalization, ensuring consistent noise suppression. To achieve precise switching of filtering modes and low signal processing latency, the frequency detection link and timing control logic are as follows:

[0049] Frequency detection hardware and interface: A high-precision frequency sensor (model: SICKFTMg) is used to collect the power grid frequency. The sensor has a measurement range of 45Hz to 55Hz, a measurement accuracy of ±0.01Hz, and a sampling interval of 10ms. The sensor is connected to the signal processing module (such as STM32H743) via an RS485 interface with a communication baud rate of 9600bps. The data frame format is defined as: start bit (1 bit) + frequency value (16 bits, decimal) + parity bit (1 bit, even parity) + stop bit (1 bit). One frame of frequency data is transmitted to the signal processing module every 10ms.

[0050] Filtering timing control logic: Set a fixed signal processing timing chain and execute the following steps sequentially:

[0051] 1. Anti-aliasing filter (processing time 10ms, using 4th order Butterworth low-pass filter).

[0052] 2. Frequency determination (takes 1ms, compares the current frequency with the ±0.5Hz threshold);

[0053] 3. Filtering mode switching (duration not greater than 50ms; if the frequency fluctuation is not greater than ±0.5Hz, wavelet packet decomposition is enabled; otherwise, digital notch filtering is switched).

[0054] 4. Normalization processing (takes 5ms, using linear normalization algorithm);

[0055] Timing control is achieved through timer interrupts in the signal processing module. The interrupt priority is set as follows: frequency judgment interrupt (highest) > switching interrupt > normalization interrupt, ensuring that the total processing delay is no more than 70ms. During the switching process, a signal buffering and continuation mechanism is enabled to buffer the filtered signal of the first 10ms to avoid data loss during the switching interval.

[0056] The frequency mapping relationship of wavelet packet decomposition is as follows: the frequency decomposition levels corresponding to scale parameter 6 are clearly defined—the first layer covers 500 to 250 Hz, the second layer 250 to 125 Hz, the third layer 125 to 62.5 Hz, the fourth layer 62.5 to 31.25 Hz, the fifth layer 31.25 to 15.625 Hz, and the sixth layer 15.625 to 7.8125 Hz; among them, the third to fifth layers (125 to 7.8125 Hz) are the main frequency band of the lifting load signal, and the sixth layer (not greater than 7.8125 Hz) is temperature drift noise. After decomposition, the signals of the third to fifth layers are retained to reconstruct the load waveform, and low-frequency noise is further removed.

[0057] Adaptive notch filter algorithm: Employing the LMS (Least Mean Square) algorithm with a step size factor μ=0.001, the reference signal is the power grid synchronous sampling signal. The notch center frequency is iteratively adjusted to achieve a tracking accuracy of ±0.1Hz. The stopband attenuation of this digital notch filter is set to -40dB to ensure that the 50Hz fundamental residual component is less than 1mV, the passband ripple is controlled within ±0.1dB, and the load ripple component from 0.1Hz to 500kHz is retained for subsequent fault analysis.

[0058] The cloud processing module adopts a distributed architecture, enabling dynamic adjustment of computing resources for each node based on real-time power grid performance and load changes. A node is marked as high-load when its CPU utilization exceeds 80%, or its IO wait time exceeds 20ms (SSD storage) or 50ms (HDD storage). When a node's load exceeds 1.5 times the cluster's average load, it is marked as overloaded and resource scheduling is triggered. The cloud processing module's resource scheduling employs a dynamic weighting algorithm, reducing the number of query subtasks dispatched to high-load nodes by 50% and increasing the subtask handling ratio of idle nodes to 80%.

[0059] The cloud processing module also uses pre-trained general image recognition model parameters, performing specialized training only on the fully connected layers of the model for the load prediction task. The pre-trained general image recognition model is either a ResNet or MobileNet model, using the ResNet-34 model (balancing accuracy and computational power), removing the original fully connected layers and replacing them with a 256-dimensional hidden layer + a 1-dimensional output layer, and using linear activation (adapted to the regression task of load prediction).

[0060] The annotation process employs a three-stage process: coarse screening, fine labeling, and validation. It includes three types of samples (50,000 samples with normal load, 20,000 samples with sudden load surges, and 30,000 samples with faulty loads). The annotation content includes load value + timestamp + fault type. A sample dataset is provided. For specialized training, the Adam optimization algorithm is used with a learning rate of 0.001, a batch size of 64, and 10,000 training iterations. The average cluster load is calculated using a 5-minute sliding window: Average Load = (CPU Utilization × 0.4 + Memory Usage × 0.3 + IO Pressure × 0.3). Idle nodes are defined as those with a load less than 50% of the average cluster load and a scheduling latency of no more than 100ms. This transfer learning approach significantly reduces the training data requirements and improves the model's generalization ability.

[0061] The model training convergence and early stopping mechanism employs a dual-index convergence control to monitor the training set loss and validation set R² (coefficient of determination) in real time during training. When the training set loss is no greater than 0.005 and the validation set R² is no less than 0.98, the model enters the observation period. If the validation set loss decreases by less than 0.001 during the observation period (100 consecutive epochs), the early stopping mechanism is triggered, and training is terminated. The model weights are saved every 200 epochs during training, and the weight file with the highest validation set R² is selected as the deployment model to avoid overfitting.

[0062] Dataset calibration and augmentation methods: The original dataset was constructed using physical calibration and simulation supplementation. 1. Physical calibration: 50,000 normal load data were collected by loading 2 to 10 tons of standard weights (0.5 tons interval) with a 10-ton lifting device; 20,000 sudden load data were collected by instantaneous loading (5 to 10 tons, 0.5 seconds) using an electromagnetic chuck. 2. Simulation supplementation: 30,000 fault data such as ripple anomaly (500kHz interference) and temperature drift (-10 to 50℃) were generated using MATLAB. Dataset augmentation adopted: time axis stretching (±10%) + Gaussian noise injection (signal-to-noise ratio 30dB). After augmentation, the total data volume reached 200,000 data, which were divided into training set, validation set and test set in a 7:2:1 ratio.

[0063] The derivation of resource scheduling weights is based on correlation analysis of 1000 hours of spreader load test data, specifically calculated using Pearson coefficients. The correlation coefficient between CPU utilization and model inference latency is 0.82 (strong correlation), memory utilization is 0.65 (moderate correlation), and IO wait time is 0.68 (moderate correlation). The weights are determined using the Analytic Hierarchy Process (AHP): a target layer (minimize inference latency) - criterion layer (CPU / memory / IO) - scheme layer (node ​​load) model is constructed. The consistency check CR=0.07 is less than 0.1, and the final weights are determined as CPU 0.4, memory 0.3, and IO 0.3 to ensure that the scheduling strategy is scientific and reasonable.

[0064] The load early warning module normalizes the real-time collected load data to unify the dimensions to the range of 0 to 100%, and generates upper and lower boundaries of the curve based on the historical prediction error range to form a dynamic confidence interval. The dynamic confidence interval adopts a large-sample normal distribution model, and the calculation formula is as follows:

[0065] Confidence interval = predicted load value ± k × σ;

[0066] Where k is the confidence coefficient (k=1.96 at 95% confidence level), and σ is the standard deviation of historical prediction error within the context window (15 / 30 minutes); an adaptive prediction fault tolerance space is constructed, which automatically reduces the deterministic weight of the prediction result when there are sudden environmental changes or abnormal loads. The weight adjustment rule is as follows: when the real-time load exceeds the upper limit of the confidence interval by 10%, the weight of the prediction result is reduced from 1.0 to 0.6, while the weight of the real-time data is increased to 0.4; after the anomaly is resolved, the weight is restored at a rate of 0.1 / minute.

[0067] Confidence interval generation during the initialization phase: When the system starts up for the first time and there is no historical data, the theoretical model and default parameters are used to generate the initial confidence interval. The initial fluctuation range is set to ±15% of the rated value based on the rated load of the spreader (e.g., 10t).

[0068] Default σ_initial = rated load × 5% (σ_initial = 0.5t for a 10t lifting device);

[0069] The confidence coefficient k=1.96; for the first 2 hours of system operation, σ is updated every 10 minutes based on real-time data acquisition. After 2 hours, the system switches to a regular rolling update mechanism with a 15 / 30 minute context window to ensure the effectiveness of early warning during the initialization phase.

[0070] The quantitative basis for the abnormal threshold includes: exceeding the upper limit of the confidence interval by 10%: refer to the yield strength of the lifting beam material (Q235 steel). For example, the yield strength of Q235 steel is 235MPa, and the design safety factor of the 10t lifting beam is 1.2. Exceeding the upper limit by 10% corresponds to a 10% redundancy of the yield strength to avoid structural overload.

[0071] Abnormal resolution lasts for 30 seconds: Based on the dynamic characteristics test of the lifting mechanism of the spreader, it was found that after a sudden increase in load, it takes 25 to 30 seconds for the vibration of the lifting mechanism to decay to a stable range of ±2%. The 30-second delay can avoid misjudgment caused by vibration.

[0072] Weight recovery rate of 0.1 / minute: Matches the convergence speed of the load prediction model. In actual tests, when the model weights recovered from 0.6 to 1.0, it took 10 minutes for the prediction error to stabilize. A rate of 0.1 / minute can ensure that the prediction accuracy gradually improves during the weight recovery process.

[0073] The context window size for the dynamic confidence interval is 30 minutes. Normalization uses a linear normalization algorithm to map load data of different ranges to a range of 0 to 100%. The calculation formula is as follows:

[0074] ;

[0075] This processing method allows data from different workloads to be compared and analyzed under a unified standard.

[0076] The execution control module wakes up the dormant power module based on the warning signal output by the load warning module. When there is a sudden increase in load or abnormal ripple, the weight of the abnormal item is increased to 50%. The Time-Sensitive Networking (TSN) protocol (IEEE 802.1Qbv) is used, with a transmission period of 10ms. The scheduling mechanism prioritizes periodic load data, ensuring a single-trip delay of no more than 30ms. Power drive and pausing control:

[0077] The power module's wake-up and drive circuit employs a two-stage drive circuit consisting of pre-wake-up and rated output stages, specifically including:

[0078] Pre-wake-up phase: The execution control module (such as PLCS7-1200) outputs a PWM signal (duty cycle 30%, frequency 1kHz) to control the MOSFET (model IRF540) to drive the power module. Within 100ms, the power module outputs 30% of the rated power (4.5 to 6W) to complete capacitor charging.

[0079] Rated output stage: After pre-wake-up, the circuit switches to a linear regulator circuit. Through PID adjustment (proportional coefficient 0.8, integral coefficient 0.2, derivative coefficient 0.1), the power is linearly increased to the rated power (15 to 20W) within 500ms. A 0.1Ω sampling resistor is connected in series in the circuit to collect the output current in real time and feed it back to the execution control module to realize overcurrent protection (overcurrent threshold 1.5A).

[0080] The calculation model for the slow-stop acceleration of the spreader is as follows: Based on the dynamic equation of the spreader, the slow-stop parameter is derived—the slow-stop acceleration system, where: system is the braking force of the hoisting mechanism (taken as 80% of the rated force of the motor, system for a 10t spreader), m is the total mass of the spreader and the load (m=11000kg when the 10t spreader is fully loaded, including 100kg of hook), g=9.8m / s2; Substituting into the calculation, a=0.5m / s2 when the 10t spreader is fully loaded, and a=0.8m / s2 when unloaded (m=100kg), the software is used to correct a to ensure a smooth stop under different loads and avoid spreader swaying.

[0081] TSN Protocol Configuration Details: The TSN network is configured using the IEEE 802.1Qbv standard. 1. Time Window: Each 10ms period is divided into cycles. The first 3ms of the cycle are allocated to load warning signals (highest priority, PCP=7), and the last 7ms are allocated to regular data (such as status monitoring). 2. Time Synchronization: The IEEE 1588PTP protocol is enabled, with the cloud processing module as the master time node and the execution control module as the slave node. The synchronization error is no greater than 1ms. 3. Flow Control: Credit-Based Shaper (CBS) is used to ensure that the warning signal transmission is packet-free and the one-way delay is no greater than 30ms.

[0082] The remaining indicators are weighted proportionally, with the following weight allocation logic: initial weights are 40% for load, 20% for ripple, 20% for temperature, and 20% for voltage. When the load suddenly increases (not less than 10% / s of the rated value) or the ripple is abnormal (greater than 500kHz), the weight of the abnormal item increases to 50%, and the remaining indicators are compressed by multiplying their original weights by 0.5 (e.g., the voltage weight drops to 10%). The standby power consumption of the power module is 5W, and the operating power consumption is 20W. When the load warning module outputs an overload warning signal (load reaches 80% to 90% of the rated value), the execution control module wakes up 1 to 2 dormant power modules in advance; when it outputs an emergency stop warning signal (load reaches 100% of the rated value), the execution control module triggers the slow-stop control of the hoisting mechanism. This graded response mechanism effectively avoids system overload and sudden failures.

[0083] The data storage module stores voltage waveforms, load responses, and environmental data from a historical fault database. Analysis of this historical data optimizes the accuracy and real-time performance of load warnings. The module uses a solid-state drive (SSD) with a capacity of 512GB and a data read speed of no more than 0.1ms. It also features incremental data backup, backing up newly added load data, warning records, and fault information to a remote server every morning to ensure data security and traceability.

[0084] Database configuration and fault tolerance mechanisms:

[0085] InfluxDB Time Series Database Optimization Configuration: 1. Retention Policy (RP): Create 3 RPs, specifically: normal_data (retained for 6 months, sharded every 1 hour, used to store normal load data), fault_data (retained for 3 years, sharded every 6 hours, used to store fault data), and env_data (retained for 1 year, sharded every 12 hours, used to store environment parameters); 2. Index Configuration: Enable Time Series Index (TSI), set the memory cache to 40% of the storage capacity (204.8GB cache for 512GB SSD) to accelerate queries; 3. Query Optimization: Adopt a tag filtering priority strategy, filtering the hangar ID and timestamp tag first, and then extracting the numerical fields to ensure that the response time of a single query is no more than 10ms.

[0086] Backup fault tolerance and alarm mechanism: Adopting a local caching and dual-path remote backup architecture, specifically including: 1. Local caching: Before triggering incremental backup at 2:00 AM every day, newly added data (within 24 hours) is cached to the backup partition of the local SSD (capacity 100GB), and a checksum (MD5) is generated; 2. Remote backup: Backup files are simultaneously transferred to two remote servers in different locations, using SFTP protocol for encrypted transmission (AES-256); 3. Fault tolerance: If the backup timeout exceeds 10 minutes or the checksum does not match, a local audible and visual alarm is immediately triggered (red LED light is constantly on + intermittent buzzer alarm, frequency 1Hz), and the local cached data is retained for 72 hours. After the fault is repaired, the backup is restarted to avoid data loss.

[0087] Through the coordinated operation of the aforementioned modules, this system achieves real-time monitoring, early warning, and control of the load on lithium battery lifting devices, significantly improving the safety and reliability of the devices. The system adopts a cloud-based processing architecture, offering high scalability and flexibility, and can adapt to load monitoring needs under various complex working conditions.

[0088] Example 2

[0089] This invention provides a method for managing a dynamic load monitoring lithium battery lifting device, specifically including the following steps:

[0090] Step 1: Data Collection

[0091] The system employs an architecture comprised of a data acquisition unit and a cloud-based system. A 1kHz high-sampling-rate ADC is used to acquire load data, which is then transmitted over a network to the cloud processing module. To improve data accuracy and real-time performance, a high-speed differential probe with a 1MHz sampling rate is used to capture the load signal throughout its entire lifecycle, ensuring the capture of minute load changes. The 16-bit sampling accuracy effectively captures even minute load variations.

[0092] Step 2: Signal Processing

[0093] The acquired load data was filtered using an anti-aliasing filter with a cutoff frequency set to 40% of the sampling rate, i.e., 400Hz. This 400Hz cutoff frequency corresponds to 40% of the ADC's 1kHz sampling rate, which is lower than the Nyquist frequency of 500Hz, effectively suppressing high-frequency interference in the 800Hz to 1MHz range. This complements the signal acquisition frequency band of the high-speed differential probe. Simultaneously, a digital notch filter was used to eliminate the 50Hz sinusoidal fundamental component. The filter design employed an adaptive algorithm to track grid frequency fluctuations in real time, dynamically adjusting the notch center frequency and setting a -40dB stopband attenuation to ensure the residual fundamental component is less than 1mV, passband ripple is controlled within ±0.1dB, and ripple components from 0.1Hz to 500kHz are retained. The filtered signal contains only ripple and noise components, providing a clean data source for subsequent analysis.

[0094] Step 3: Cloud Processing

[0095] A distributed architecture is adopted, dynamically adjusting the computing resources of each node based on the real-time changes in the power grid and load. Pre-trained general image recognition model parameters are used, with specialized training performed only on the fully connected layer for load prediction, significantly reducing the need for labeled data. During training, the SGD optimization algorithm is used with a learning rate of 0.0001, a batch size of 128, and 20,000 training iterations.

[0096] Step 4: Load Alert

[0097] The real-time collected load data is normalized to a uniform dimension of 0 to 100%. Based on historical prediction error ranges, upper and lower boundaries of the curve are generated to form a dynamic confidence interval, constructing an adaptive prediction tolerance space. When environmental changes or load anomalies occur, the system automatically reduces the deterministic weight of the prediction results. The context window size of the prediction tolerance space is 15 minutes.

[0098] Step 5: Execution Control

[0099] Based on the warning signal output by the load warning module, the power module is promptly woken up from hibernation. If there is a sudden increase in load or abnormal ripple, the weight of the abnormal item is increased to 50%, while the weights of other indicators are proportionally reduced to enhance sensitivity to hardware failures. The power module's standby power consumption is 3W, and its operating power consumption is 15W.

[0100] Step Six: Data Storage

[0101] The system stores voltage waveforms, load responses, and environmental data from a historical fault database. Analysis and mining of this data further optimizes the accuracy and real-time performance of load warnings. Data storage utilizes solid-state drives (SSDs) with a capacity of 1TB and a read speed of 0.05ms.

[0102] Through the above steps, this method enables real-time monitoring and early warning of the load on lithium battery lifting devices. It can effectively capture minute load change signals, filter out noise interference, and improve computing efficiency through cloud processing, ultimately achieving accurate load early warning and control, thereby improving the safety and reliability of lithium battery lifting devices.

[0103] Example 3

[0104] This invention provides a cloud-based dynamic load monitoring and lithium battery lifting device management method, which includes the following steps:

[0105] Step 1: Data Acquisition. An architecture consisting of a data acquisition unit and a cloud platform is employed. Load data is acquired using a 1kHz high-sampling-rate ADC and transmitted over the network to the cloud processing module. To improve data accuracy and real-time performance, a high-speed differential probe with a 1MHz sampling rate is used to capture the load signal throughout its entire lifecycle, ensuring the capture of minute load changes. The sampling precision is 16 bits, effectively capturing even minute load variations.

[0106] Step Two: Signal Processing. The acquired load data is filtered using an anti-aliasing filter. The cutoff frequency is set to 40% of the sampling rate, i.e., 400Hz. This 400Hz cutoff frequency corresponds to 40% of the ADC's 1kHz sampling rate, which is lower than the Nyquist frequency of 500Hz, effectively suppressing high-frequency interference in the 800Hz to 1MHz range. This complements the signal acquisition frequency band of the high-speed differential probe. Simultaneously, a digital notch filter is used to eliminate the 50Hz sinusoidal fundamental component. The filter design employs an adaptive algorithm, dynamically adjusting the notch center frequency based on real-time grid frequency fluctuations. A -40dB stopband attenuation is set to ensure the residual fundamental component is less than 1mV, passband ripple is controlled within ±0.1dB, and ripple components from 0.1Hz to 500kHz are retained. The filtered signal contains only ripple and noise components, providing a clean data source for subsequent analysis.

[0107] Step 3: Cloud Processing. A distributed architecture is adopted, dynamically adjusting the computing resources of each node based on the real-time status of the power grid and load changes. Pre-trained general image recognition model parameters are used, with specialized training performed only on the fully connected layers for load prediction, significantly reducing the need for labeled data. During training, the SGD optimization algorithm is used with a learning rate of 0.0001, a batch size of 128, and 20,000 training iterations.

[0108] Step 4: Load Warning. The real-time collected load data is normalized, unifying the units to the 0-100% range. Based on historical prediction error ranges, upper and lower boundaries of the curve are generated, forming a dynamic confidence interval and constructing an adaptive prediction tolerance space. When environmental changes or load anomalies occur, the system automatically reduces the deterministic weight of the prediction results. The context window size of the prediction tolerance space is 15 minutes.

[0109] Step 5: Execution Control. Based on the warning signal output from the load warning step, promptly wake up the dormant power modules. If there is a sudden increase in load or abnormal ripple, increase the weight of the abnormal item to 50%, and reduce the weight of other indicators proportionally to enhance sensitivity to hardware failures. The standby power consumption of the power module is 3W, and the operating power consumption is 15W.

[0110] Step Six: Data Storage. Store voltage waveforms, load responses, and environmental data from the historical fault database. Analyze and mine this data to further optimize the accuracy and real-time performance of load warnings. Data storage uses solid-state drives (SSDs) with a capacity of 1TB and a read speed of 0.05ms.

[0111] In a preferred embodiment, wavelet packet decomposition can replace digital notch filters for high-frequency noise stripping, effectively suppressing electromagnetic interference from contaminating the detection system. The scale parameter of wavelet packet decomposition is 6, which effectively extracts the transient features of the load signal.

[0112] In another preferred embodiment, the training process may also use the Adam optimization algorithm with a learning rate of 0.001, a batch size of 64, and 10,000 training iterations.

[0113] In yet another preferred embodiment, the context window size for predicting the fault tolerance space can be set to 30 minutes to accommodate different load variation characteristics.

[0114] In a further preferred embodiment, the power module can have a standby power consumption of 5W and an operating power consumption of 20W to meet different power requirements.

[0115] In another preferred embodiment, the data storage capacity can be 512GB and the data access speed is 0.1ms, which is suitable for smaller-scale monitoring systems.

[0116] Experimental verification shows that using the above method for dynamic load monitoring can achieve early warning of load anomalies. The system's response time to load surges is no more than 100ms, which is more than 50% faster than existing systems (100-200ms). For slow overloads (such as increasing from 80% to 100% of rated load within 10 minutes), a pre-warning can be issued 15 minutes in advance, with an accuracy rate of over 95%. Data storage uses a time-series database (InfluxDB) with a composite index of timestamp + hoist ID, and the query response time is no more than 10ms. The backup mechanism consists of local SSD caching + incremental backup on a remote server. When backup fails, a local alarm is triggered and a 72-hour cache is retained.

[0117] Experimental verification details:

[0118] Test object: 10-ton lithium battery lifting device (rated load 10t, overload limit 12t);

[0119] Test environment: Temperature -10 to 50℃, humidity 30% to 80%;

[0120] Verification instruments: high-precision force gauge (accuracy ±0.1%FS), oscilloscope (sampling rate 10MHz);

[0121] Experimental data were calibrated using reference instruments to ensure accuracy.

[0122] Load data calibration: Connect a high-precision force gauge (model HBMC16, range 0 to 20t, accuracy ±0.1%FS) in series between the hook of the lifting device and the weight. Collect the force gauge reading every 100ms and compare it with the load data collected by the system FBG sensor to calculate the deviation. If the deviation is greater than ±0.2%, adjust the temperature compensation parameter of the FBG sensor to ensure that the load measurement error is not greater than ±0.1%FS.

[0123] Signal synchronization calibration: Using an oscilloscope (model Tektronix MSO464), the power grid frequency signal (from the SICKFTMg sensor), the load signal (from the ADC), and the warning signal (from the execution control module) are simultaneously acquired. By using the oscilloscope's trigger synchronization function (trigger source is the power grid frequency signal), the timestamp deviation of the three signals is ensured to be no more than 1ms, and the filter switching delay and warning response time are verified.

[0124] Data recording: 1000 hours of continuous testing, including 200 load surges (5→10t, time 0.5s) and 50 fault simulations (ripple abnormality, temperature drift).

[0125] The experimental operation adopted a combination of automated control and manual calibration.

[0126] Load surge test: The electromagnetic chuck (with a suction force of 15t) is controlled by a PLC (Siemens S7-1200). First, a 10t weight is placed on the ground. After the weight is connected to the hook of the lifting device, the PLC sends an instantaneous loading signal. At the same time, the electromagnetic chuck attracts the weight, causing the load to surge from 5t (the weight of the lifting device) to 10t in 0.5s. After each surge, the device is left to stand still for 5 minutes. This is repeated 200 times, and the load data and warning signals are recorded simultaneously.

[0127] Fault simulation test: 1. Ripple abnormality: A 500kHz, 1Vpp sinusoidal interference signal was injected into the power line of the spreader using a signal generator (model Tektronix AFG3102) for 10 minutes to simulate ripple abnormality; 2. Temperature drift: The spreader was placed in a constant temperature chamber (model BinderMK53), and the temperature was raised from -10℃ to 50℃ at a rate of 5℃ / hour for 24 hours to simulate temperature drift; each fault was repeated 25 times, and the system response at the time of the fault was recorded.

[0128] The accuracy of the early warning is calculated using a binary classification confusion matrix, with the specific criteria as follows:

[0129] Sample definition: True positive (TP): Overload warning of 80% or 100% overload warning of 100% load exceeding the rated value, and the system outputs a warning signal within 100ms;

[0130] True negative (TN): The load is normal (not exceeding 80% of the rated value), and the system has no warning signal;

[0131] False positive (FP): The system outputs a false warning despite normal load.

[0132] False negative (FN): Abnormal load but the system does not output a warning.

[0133] Accuracy formula: Early warning accuracy = (TP+TN) / (TP+TN+FP+FN)×100%;

[0134] In the experiment, all 200 load surges were TP (without FN), all 50 fault simulations were TP, and there were no FP during the 1000-hour normal load test (TN=1000×3600 / 10=360000). The final calculated early warning accuracy rate was (250+360000) / (250+360000+0+0)×100%≈99.93%, which meets the design requirements.

[0135] The system's response time to sudden load changes is less than 100ms, effectively preventing equipment damage and safety accidents caused by abnormal loads. The system in the above embodiments, through the collaborative operation of its modules, achieves real-time monitoring, early warning, and closed-loop control of the lithium battery lifting load, significantly improving the safety and reliability of the lifting device. Adopting a cloud-based distributed architecture, it possesses high scalability and flexibility, adapting to load monitoring needs under complex working conditions.

[0136] The methods described in the above embodiments achieve real-time and accurate acquisition of the load of lithium battery lifting devices, efficient noise filtering, and efficient cloud computing through multi-step collaboration, ultimately achieving accurate load warning and control, and significantly improving the safety and reliability of lithium battery lifting devices.

[0137] The cloud processing method in the above embodiments enables real-time monitoring, accurate early warning, and flexible control of the load on the lithium battery lifting device. Different preferred embodiments can be adapted to various scenario requirements. Experimental verification shows that the early warning response is fast and the accuracy is high, effectively ensuring the safe operation of the lifting device.

[0138] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

[0139] In the description of this invention, it should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions provided in this disclosure can be achieved, and no limitation is imposed herein.

[0140] The above description is merely a preferred embodiment of the present invention and does not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A dynamic load monitoring lithium battery lifting device management system, characterized in that, The system includes: The data processing module is used for data processing in the lithium battery lifting device management system. The data acquisition module adopts a distributed architecture. It acquires load data of the lifting device through an ADC adapted to the load monitoring requirements, and simultaneously captures load signals with a high-frequency differential probe. The acquired raw data is directly transmitted to the signal processing module for preprocessing. The signal processing module performs filtering and optimization on the raw data output by the data acquisition module. It suppresses high-frequency interference through an anti-aliasing filter and removes electromagnetic noise by wavelet packet decomposition. The clean data after processing is pushed to the cloud processing module. The raw data undergoes a three-stage preprocessing process before filtering optimization. The three-stage preprocessing includes: Multi-source data synchronization calibration, environmental error correction, and raw data verification; The cloud processing module dynamically adjusts node computing resources based on clean data transmitted by the signal processing module, combined with real-time power grid status and load changes. At the same time, it uses a general image recognition model to achieve model recognition through pre-training, and achieves load prediction through a fully connected layer of a specially trained model to generate prediction results, which are then fed back to the load early warning module. The load early warning module receives the prediction results from the cloud processing module, normalizes the real-time load data to unify the units, and generates a dynamic confidence interval and adaptive fault tolerance space by combining the historical error data in the data storage module. When an anomaly is detected, the weight of the prediction result is adjusted and the execution control module is triggered. The execution control module includes a power module. The execution control module is used to wake up the dormant power module in response to the warning signal from the load warning module. When a sudden increase in load or abnormal ripple is detected, the weight ratio of the abnormal item is increased, and the weights of other indicators are adjusted proportionally to ensure that the system responds to key risks first. The data storage module stores historical fault data. By mining historical data, it provides data support for the load early warning module to optimize the confidence interval and for the cloud processing module to improve prediction accuracy, thus forming a closed loop of system iteration.

2. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The data acquisition module is also associated with an FBG sensor, which is used to acquire wavelength drift data of the spreader. The system also includes a timing module, which is communicatively connected to the data acquisition module to achieve time synchronization and alignment of the load data acquired by the ADC, the load signal captured by the high-frequency differential probe, and the wavelength drift data acquired by the FBG sensor.

3. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The signal processing module performs a three-level preprocessing procedure on the raw data, including multi-source data synchronous calibration, environmental error correction, and raw data verification. The signal processing module also includes a frequency judgment unit, which is used to switch the filtering mode according to the power grid frequency fluctuation. When the power grid frequency fluctuation is no greater than ±0.5Hz, wavelet packet decomposition is enabled, and when the power grid frequency fluctuation is greater than ±0.5Hz, it switches to a digital notch filter.

4. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The cloud processing module determines the node load status based on node CPU utilization, memory usage, and IO wait time. When the node CPU utilization exceeds a preset CPU load threshold, or the IO wait time exceeds a preset first IO threshold when the node uses SSD storage or a preset second IO threshold when the node uses HDD storage, the node is marked as a high-load node. When the node load exceeds a preset multiple threshold of the cluster average load, resource scheduling is triggered. The preset CPU load threshold is no less than 80%, the preset first IO threshold is no less than 20ms, the preset second IO threshold is no less than 50ms, and the preset multiplier threshold is no less than 1.5 times; the pre-trained general image recognition model is a ResNet model or a MobileNet model, and the cloud processing module only performs load prediction task-specific training on the fully connected layers of the model.

5. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The load early warning module generates dynamic confidence intervals based on a large-sample normal distribution model. The formula for calculating the dynamic confidence interval is as follows: Confidence interval = predicted load value ± k × σ; Where k is the confidence coefficient and σ is the standard deviation of historical prediction error within the set context window; When the real-time load exceeds the confidence interval, the load warning module lowers the weight of the prediction result and simultaneously increases the weight of the real-time data.

6. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The execution control module transmits warning signals via the TSN protocol and allocates the first 3ms of the cycle to the load warning signal for priority transmission. The execution control module is also used to combine the dynamic parameters of the spreader to realize the slow stop control of the spreader, and dynamically adjust the slow stop acceleration according to the total mass of the spreader and the load.

7. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The data storage module adopts a time-series database, which is configured with at least three retention strategies: a first retention strategy for storing normal load data, a second retention strategy for storing fault data, and a third retention strategy for storing environmental parameters. The data storage module also adopts a multi-path backup architecture, which achieves data backup through local caching and remote backup in collaboration. When the backup process meets preset abnormal conditions, it triggers local alarms and preset fault tolerance processing. The preset abnormal conditions include at least one of backup timeout and backup data inconsistency, and the preset fault tolerance processing includes at least one of retaining local backup cache and re-initiating backup after fault repair.

8. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The system also includes an FBG protection module, which includes a stainless steel protective shell, a silicone buffer pad, and an oil-resistant shielded cable. The stainless steel protective shell is fitted over the FBG sensor, the silicone buffer pad is filled between the stainless steel protective shell and the FBG sensor, the oil-resistant shielded cable is connected to the lead wire of the FBG sensor, and the cable joint is sealed with waterproof sealant.

9. The dynamic load monitoring lithium battery lifting device management system according to claim 1 or 2, characterized in that, The system also includes a frequency sensing module, which is used to collect power grid frequency data and transmit it to the signal processing module; The frequency sensing module's measurement range is adapted to the power grid frequency fluctuation range, the sampling interval meets the filtering mode switching timing requirements of the signal processing module, and it establishes a connection with the signal processing module through a serial communication interface that supports stable data transmission.

10. The dynamic load monitoring lithium battery lifting device management system according to claim 1, characterized in that, The power module of the execution control module adopts the following two-stage driving method: During the pre-wake-up phase, the execution control module outputs a PWM signal to drive the power module, enabling the power module to output a preset pre-charge power within a preset pre-charge time, thus completing the pre-charge of the power module to avoid startup impact. During the rated output phase, the execution control module switches to a voltage regulator circuit that can achieve stable power output, and uses a closed-loop regulation method to enable the power module to linearly increase to the rated power within a preset rated power increase time.

Images