Composite material container real-time state early warning system based on multi-mode perception

Abstract

The invention relates to the technical field of container monitoring, and discloses a composite material container real-time state early warning system based on multi-mode perception. The system comprises a multi-mode sensing unit, a real-time acceleration processing unit, an anomaly analysis unit and an early warning output unit. The multi-modal sensing unit collects multi-dimensional real-time data such as vibration, temperature, humidity, pressure and strain, and comprehensively captures the structural state, environmental influence and load condition of the container. The real-time acceleration processing unit is used for quickly converting the acquired data into state feature vectors through matrix multiplication and addition operation so as to realize real-time data processing; the anomaly analysis unit carries out multi-dimensional anomaly evaluation based on the state feature vector, identifies anomaly in combination with multi-modal data relevance, and avoids single threshold judgment limitation; and the early warning output unit generates an early warning signal when detecting abnormity and feeds back information in time. The system can carry out comprehensive, real-time and accurate monitoring and early warning on the state of the composite material container, and is suitable for full-life-cycle state management of the composite material container.

Description

Technical Field

[0001] This invention relates to the field of container monitoring technology, specifically to a real-time status early warning system for composite material containers based on multimodal perception. Background Technology

[0002] In modern logistics and transportation systems, containers serve as the core carriers for cargo storage and transshipment, and their performance stability directly impacts cargo transportation safety and supply chain efficiency. With advancements in materials science, composite materials, leveraging their advantages such as lightweight, corrosion resistance, and fatigue resistance, are gradually replacing traditional metal materials in container manufacturing. However, during long-term use, composite material containers are susceptible to problems such as micro-damage to the internal structure, surface cracking, and decreased sealing performance due to the complex and ever-changing transportation environment, including bumps and vibrations, alternating temperature and humidity changes, external impacts, and stress concentration during cargo loading. These problems are often difficult to detect through manual inspections in their early stages. Once accumulated and spread, they can lead to reduced container load-bearing capacity, moisture damage to cargo, and even safety accidents during transportation, resulting in significant economic losses for logistics companies. Currently, technical solutions for container condition monitoring have significant limitations. Traditional manual inspection methods rely on the experience and judgment of staff, which is not only inefficient but also difficult to detect microscopic damage within composite materials, posing a significant risk of missed detections. Many existing automated monitoring systems use single-modal sensors, such as vibration or temperature sensors, which cannot comprehensively reflect the complex condition of the container. For example, vibration data alone can only determine the impact intensity during transportation but cannot identify internal aging of composite materials caused by humidity changes; pressure sensors alone cannot detect minute cracks on the container surface. This single-dimensional monitoring approach easily leads to incomplete and missing data, resulting in misjudgments or missed detections of abnormal conditions. The data processing capabilities of existing monitoring systems are also insufficient. Containers generate a large amount of real-time monitoring data during transportation, but traditional data processing units are slow and struggle to perform real-time analysis and processing of multi-source data. Data often needs to be transmitted to remote servers for further processing, which not only increases data transmission latency but also risks data loss due to network fluctuations, preventing the timely generation of anomaly warning signals. Furthermore, in the anomaly assessment stage, existing systems often employ a simple threshold method, setting fixed numerical thresholds and classifying data as abnormal when the threshold is exceeded. This assessment method fails to consider the correlation and coupling between multimodal data, making it difficult to handle the multi-dimensional anomalies of composite material containers under complex operating conditions. The assessment accuracy is low, failing to meet the demands of modern logistics for real-time and accurate monitoring of container status. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time status early warning system for composite material containers based on multimodal perception, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a real-time status early warning system for composite material containers based on multimodal perception, the system comprising: A multimodal sensing unit is used to collect real-time multimodal sensor data from composite material containers. A real-time acceleration processing unit is used to perform matrix multiplication and addition operations on the data collected by the multimodal sensing unit to generate a state feature vector; An anomaly analysis unit is used to perform multidimensional anomaly assessment based on the state feature vector; The early warning output unit is used to generate an early warning signal when the anomaly analysis unit detects an anomaly.

[0005] Preferably, the real-time acceleration processing unit includes: Multi-source signal generation module, used to generate multiple parallel signal sources; A weight loading module is used to load weights onto the signals generated by the multi-source signal generation module according to a preset weight vector. The signal modulation module is used to modulate the data collected by the multimodal sensing unit onto the signal after the weight loading module has loaded the weights; The time delay module is used to apply a time delay to the modulated signal; The feature segment extraction module is used to extract feature segments within a specific time window from the delayed signal; The signal superposition module is used to linearly superimpose the extracted feature segments.

[0006] Preferably, the real-time acceleration processing unit further includes a dynamic state modeling module; The dynamic state modeling module is used to construct a dynamic state model based on historical state data, and takes the output of the signal superposition module as input to generate theoretical state values.

[0007] Preferably, the anomaly analysis unit includes a multidimensional difference calculation module; The multidimensional difference calculation module is used to perform multidimensional difference analysis on the theoretical state values ​​and measured state values ​​generated by the dynamic state modeling module, and generate a difference coefficient matrix. The multidimensional difference analysis includes time-domain cumulative deviation calculation, frequency-domain energy shift detection, and signal sequence similarity evaluation.

[0008] Preferably, the anomaly analysis unit further includes a spatial positioning module; The spatial positioning module is used to input the difference coefficient matrix generated by the multidimensional difference calculation module into the spatial topology model, generate an anomaly probability distribution map, and locate the abnormal physical region.

[0009] Preferably, the spatial positioning module includes: The topology modeling submodule is used to construct the spatial topology of composite container; An anomaly propagation simulation submodule is used to simulate anomaly propagation paths based on the difference coefficient matrix. The probability distribution generation submodule is used to generate anomaly probability distribution maps covering the entire area.

[0010] Preferably, the anomaly analysis unit further includes an adaptive learning module; The adaptive learning module is used to update the model parameters of the dynamic state modeling module based on the anomaly probability distribution map; The adaptive learning module includes: an activation neuron identification submodule, used to identify the activation neurons of the current data point; The fuzzy neuron determination submodule is used to determine fuzzy neurons based on the intersection relationship between reference data points and activated neurons; The learning rate calculation submodule is used to calculate the learning rate of the fuzzy neuron.

[0011] Preferably, the fuzzy neuron determination submodule includes a jump data point analysis unit; The jump data point analysis unit is used to calculate the probability of anomalies in data points and determine the target jump data point based on the probability of anomalies. The fuzzy neuron identification submodule is used to identify fuzzy neurons based on target jump data points.

[0012] Preferably, the learning rate calculation submodule includes a position change evaluation unit; The position change evaluation unit is used to calculate the variance of neuron position change; The learning rate calculation submodule is used to calculate the learning rate of the fuzzy neuron based on the position variation variance and the preset learning rate of the activated neuron.

[0013] Preferably, the early warning output unit includes an anomaly decision module; The anomaly decision-making module is used to generate early warning strategies based on the anomaly probability distribution map and the updated model parameters; The early warning strategy includes activating high-frequency monitoring mode and executing physical disturbance tests.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By acquiring real-time multimodal sensor data from containers through multimodal sensing units, comprehensive information on various states of the containers during use can be captured. Compared to traditional single-modal monitoring methods, multimodal data covers multiple dimensions such as vibration, temperature, humidity, pressure, and strain, comprehensively reflecting the container's structural state, environmental influences, and load conditions. This avoids misjudgments or omissions due to incomplete data, making container status monitoring more comprehensive and complete. Whether it's internal microscopic damage to composite material containers, performance fluctuations caused by environmental changes, or localized stress anomalies caused by improper cargo loading, all can be effectively captured through the collaborative acquisition of multimodal data, providing rich and comprehensive basic information for subsequent status analysis. The real-time acceleration processing unit performs matrix multiplication and addition operations on the data collected by the multimodal sensing unit to generate state feature vectors, significantly improving the efficiency and speed of data processing. During container transportation, monitoring data is continuously and massively generated. Traditional data processing units struggle to handle this high-concurrency data processing demand, often resulting in data backlog and analysis delays. However, the real-time acceleration processing unit, with its efficient matrix multiplication and addition capabilities, can perform real-time computation and feature extraction on multi-source heterogeneous monitoring data, quickly transforming raw data into feature vectors that reflect the container's state. This eliminates the need for remote servers, effectively shortening the time interval between data collection and analysis, ensuring real-time monitoring and analysis of container states, and laying a crucial foundation for timely detection of anomalies. Even in scenarios involving long-distance transportation and continuously increasing data volumes, this processing unit can stably and efficiently complete data processing tasks, ensuring the system's dynamic tracking capability of container states. The anomaly analysis unit performs multidimensional anomaly assessment based on state feature vectors, overcoming the limitations of traditional threshold judgment methods and fully considering the correlation and coupling between multimodal data. Anomalies in composite container states are often not caused by a single factor, but rather are the result of the combined effects of multiple dimensional parameters. Traditional single-threshold judgment cannot accurately identify such multidimensionally correlated anomalies. Multidimensional anomaly assessment integrates information from multiple dimensions in the state feature vector, analyzes the mutual influence and changing patterns between different parameters, and determines whether the container is in an abnormal state holistically. For example, when vibration data shows minor anomalies, combining the trends in temperature and humidity data can accurately determine whether it is a temporary fluctuation caused by a short-term impact or a material performance degradation caused by alternating temperature and humidity. This significantly improves the accuracy and reliability of anomaly assessment, effectively reducing misjudgments and omissions of anomalies and ensuring accurate identification of potential risks to the container. The early warning output unit generates an early warning signal when the anomaly analysis unit detects an anomaly, enabling timely feedback of the container's abnormal status to relevant personnel or the control system. Timely detection and handling of container anomalies are crucial in logistics transportation. Delayed warnings can lead to further deterioration of the situation, causing more serious safety accidents or economic losses. This early warning output unit generates an early warning signal the moment an anomaly is detected. The form of the warning signal can be flexibly set according to actual needs, such as audible and visual alarms, SMS notifications, and system pop-up prompts, ensuring that relevant personnel can quickly obtain anomaly information and take timely countermeasures, such as adjusting transportation routes, suspending transportation for maintenance, and reinforcing cargo loading. This effectively curbs the spread of the anomaly, reduces the probability of cargo damage, container damage, or even transportation accidents, ensures the safety and stability of the logistics transportation process, and safeguards the economic interests of logistics companies and the smooth operation of the supply chain. The collaborative working mechanism of the entire system enables fully automated operation from data acquisition, real-time processing, multi-dimensional evaluation to early warning output, achieving real-time monitoring and anomaly warning of container status without human intervention. This automated operation mode not only reduces labor costs and avoids subjective errors and human negligence in manual operation, but also significantly improves the efficiency and stability of monitoring work. Whether in container yards at port terminals or on long-distance freight vehicles, the system can operate continuously and stably, providing strong support for the full life-cycle status management of composite material containers and promoting the development of modern logistics monitoring technology towards a more efficient, accurate, and intelligent direction. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the real-time status early warning system for composite material containers based on multimodal perception as described in this invention. Figure 2 A flowchart for signal processing in a real-time acceleration processing unit; Figure 3 A flowchart for the multidimensional difference calculation module analysis; Figure 4 A flowchart for updating parameters of the adaptive learning module. Detailed Implementation

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] Please see Figure 1The present invention provides a real-time status early warning system for composite material containers based on multimodal perception. The system includes: a multimodal perception unit, a real-time acceleration processing unit, an anomaly analysis unit, and an early warning output unit.

[0018] Multimodal sensing units are deployed in key physical areas of the composite material container to collect real-time multimodal sensor data, including strain, temperature, vibration, and acoustic emission signals. The sensor network is distributed, covering the entire container structure, and data is transmitted to the real-time acceleration processing unit via a high-speed bus. Upon receiving the multimodal sensor data, the real-time acceleration processing unit performs matrix multiplication and addition operations, converting the multi-source data stream into a unified format state feature vector. Matrix multiplication and addition operations involve multiplying the weight matrix with the input data vector and adding the bias vector, generating a high-dimensional feature representation. The state feature vector is then passed to the anomaly analysis unit, which analyzes the vector elements based on a multidimensional anomaly assessment algorithm to calculate anomaly indicators. Multidimensional anomaly assessment includes feature comparisons in the time, frequency, and spatial domains, judging abnormal states through predefined threshold rules. When an anomaly is detected, the early warning output unit generates an early warning signal, which is output to an external monitoring platform via a communication interface, triggering visual or auditory alarms and recording the timestamp and location information of the anomaly event. The entire system is implemented using an embedded hardware platform to ensure real-time performance and low power consumption.

[0019] Example 1: See Figure 2 Through a series of highly coordinated modular operations, raw multimodal sensor data is transformed into state feature vectors suitable for advanced analysis. Its implementation embodies a complete process from signal generation to feature integration, with each module designed to perform signal processing functions and ensure the parallelism and timeliness of the overall processing. The multi-source signal generation module, as the starting point of the entire processing flow, has the core function of generating multiple parallel signal sources. In practical deployments, this module is typically implemented by a programmable waveform generator capable of simultaneously generating multiple independent reference signals. The waveform types of these signal sources are configured according to the needs of subsequent processing, typically including sine waves, square waves, and pulse signals. The number of signals strictly corresponds to the number of sensor channels in the multimodal sensing unit; for example, if the system deploys sixteen sensor channels, this module will generate sixteen parallel signals. Each signal has the same initial frequency and amplitude, but its phase can be fine-tuned according to the configuration to match the physical characteristics of different sensors. The signal generation process is based on digital frequency synthesis technology, achieving high-precision waveform output through the cooperation of lookup tables and digital-to-analog converters. The generated signals are temporarily stored in a high-speed cache in the form of a digital sequence, awaiting processing by subsequent modules.

[0020] The weighting module receives parallel signal streams from the multi-source signal generation module and applies preset weighting coefficients to them. These weighting coefficients are stored in the system's non-volatile memory, forming a multi-dimensional weight vector. The dimension of the weight vector is exactly the same as the number of signal channels, and each weight value corresponds to the importance coefficient of a specific signal channel. The weighting process is implemented through a digital multiplier array, with each multiplier responsible for multiplying one signal with its corresponding weight. The weight values ​​can be fixed or dynamically adjusted according to the system's operating state. For example, during system initialization, the weight vector may use a default uniform distribution; while during operation, the system can adaptively adjust the weight allocation based on the signal-to-noise ratio characteristics of the sensor data, giving higher-quality sensor channels a larger weighting coefficient. The weighted signal not only retains the original waveform characteristics but also incorporates the system's assessment information on the importance of various signals.

[0021] The signal modulation module is responsible for fusing the actual acquired sensor data with the weighted signal. This module receives two input streams: the weighted signal from the weighting module and the multimodal sensor data from the multimodal sensing unit. The modulation process employs amplitude modulation, using the sensor data as the modulation index and multiplying it with the carrier signal. Specifically, each sensor data sample point is multiplied by the amplitude of the carrier signal at the corresponding moment, thus embedding the data information onto the carrier waveform. This modulation method maintains the dynamic range of the original data while utilizing the periodic characteristics of the carrier signal. During modulation, the system normalizes the sensor data to ensure the modulation index is within an appropriate range, avoiding signal distortion caused by over-modulation. The modulated signal retains the frequency characteristics of the carrier signal while incorporating the amplitude information of the sensor data, forming a composite signal structure.

[0022] The time delay module applies a controllable time delay to the modulated signal. This delay is set based on the propagation characteristics of the multimodal signal and the system sampling rate. Signals from different sensors may have inherent time differences due to variations in physical location or acquisition links. The time delay module uses a programmable delay line to time-align these signals. The delay can be set to a fixed value or dynamically adjusted according to the signal type. For example, a smaller delay may be needed for high-frequency vibration signals, while a larger delay can be set for low-frequency temperature signals. The delay is typically implemented using digital buffering technology, employing a first-in-first-out (FIFO) memory to time-shift the signal sequence. The delayed signals are synchronized on the time axis, providing a time-aligned data foundation for subsequent feature extraction.

[0023] The feature segment extraction module extracts feature segments within a specific time window from the time-delayed signal. The length of this time window is carefully designed to include sufficient signal period to reflect signal characteristics while maintaining sufficient timeliness to meet real-time processing requirements. The window's sliding step size is smaller than the window length, ensuring partial overlap between adjacent windows and avoiding the omission of important features. The extraction process is implemented using a sliding window algorithm, where signal segments within each window are independently extracted and subjected to feature analysis. The extracted feature segments not only contain the signal's amplitude information but also retain its time-domain waveform characteristics. These segments are temporarily stored in a feature buffer, awaiting subsequent processing.

[0024] The signal superposition module linearly superimposes the extracted feature segments. This superposition process is not a simple averaging, but a linear combination with weighted coefficients. Each feature segment is assigned a different superposition coefficient based on its source sensor type and signal quality. The superposition operation is implemented through a digital adder array, with each signal being added point-to-point while maintaining time alignment. The superimposed signal forms a composite feature representation that integrates information from multiple sensors, highlighting common features and suppressing random noise. This composite signal, as the final output of the real-time acceleration processing unit, is passed to subsequent units for further analysis.

[0025] The implementation of the entire real-time acceleration processing unit embodies a highly parallel and pipelined design philosophy. Modules are connected via a high-speed data bus, with data flowing sequentially through each processing stage in a pipelined manner. This design maximizes processing efficiency, ensuring the system can handle high-speed, multimodal sensor data. Each module is implemented using a digital signal processor or application-specific integrated circuit (ASIC) to optimize processing speed and energy efficiency. Parameter transfer and coordination between modules are managed by a central scheduling unit, ensuring the synergy and stability of the entire processing flow.

[0026] Example 2: See Figure 3By establishing a theoretical state benchmark and comparing it with real-time data from multiple dimensions, a quantitative judgment of the system's abnormal state is formed. Its implementation process integrates time series forecasting and multi-domain feature analysis techniques, realizing the transformation from raw features to abnormal indicators. The dynamic state modeling module constructs a prediction model based on historical state data. This module receives real-time state feature vectors from the signal superposition module and continuously accesses historical data sequences stored in a circular buffer. Historical data contains state feature vectors from a specific past time period (e.g., 24 hours), stored in timestamp order. The modeling process adopts an autoregressive integral moving average algorithm framework, with the model order dynamically configured according to data characteristics. Model parameters are updated online using a recursive least squares algorithm, recalculating parameter estimates each time new data is input. This module outputs theoretical state values, i.e., the current moment's state feature vector predicted based on historical patterns. The theoretical state values ​​are represented as floating-point arrays, with each element corresponding to the expected numerical range of a state feature. Prediction calculations are executed on a dedicated digital signal processor, completing a single prediction iteration every millisecond, meeting real-time requirements.

[0027] The multidimensional difference calculation module simultaneously receives two input data streams: theoretical state values ​​generated by the dynamic state modeling module and measured state values ​​collected by the multimodal sensing unit. The measured state values ​​undergo the same time alignment processing as the theoretical values ​​to ensure temporal consistency in the comparison. Difference analysis is performed in parallel across three independent dimensions: the time-domain cumulative deviation dimension calculates the cumulative difference between the theoretical and measured values ​​within a fixed time window. This calculation employs a sliding window integral algorithm, independently calculating the sum of the absolute values ​​of the differences for each state feature channel. The time window length is configurable (e.g., 500 milliseconds), and the window sliding step size is set to an integer multiple of the sampling interval. The calculation results reflect the degree of deviation of the state features in the time dimension.

[0028] Frequency domain energy shift detection transforms theoretical and measured values ​​into frequency domain analysis. This process employs a Fast Fourier Transform (FFT) algorithm to decompose the time-domain signal into spectral components. For each frequency component, the energy ratio of the theoretical spectrum to the measured spectrum is calculated, and the logarithm is taken to obtain the shift in decibels. Emphasis is placed on energy variations in characteristic frequency bands, such as the 0-100Hz band for vibration signal analysis and the 20-80kHz band for acoustic emission signal analysis. Energy shift detection can identify features such as periodic anomalies or resonant frequency changes.

[0029] The signal sequence similarity evaluation dimension employs a dynamic time warping algorithm, which overcomes the influence of small time axis offsets and calculates the minimum matching distance between the theoretical and measured sequences. The calculation process establishes a cost matrix, searches for the optimal curvature path, and finally outputs a normalized similarity coefficient. This coefficient reflects the degree of similarity in the overall waveform morphology and is capable of identifying both impulsive and gradual anomalies.

[0030] The analysis results from the three dimensions are integrated into a difference coefficient matrix, a two-dimensional data structure. Row indices correspond to time points, and column indices contain combinations of the three dimensions (time-domain cumulative deviation, frequency-domain energy shift, and signal sequence similarity) and various state feature channels. Each matrix element stores a normalized difference value, which is mapped to the [0,1] interval through a linear transformation. The matrix generation process employs a pipelined architecture, with each dimension calculated independently and then aggregated into the matrix construction unit. The final output difference coefficient matrix is ​​passed to subsequent analysis units as the basic input for anomaly localization.

[0031] The entire implementation process utilizes dedicated hardware accelerators for parallel processing. The dynamic state modeling module runs on a floating-point unit, supporting accelerated matrix operations. The multidimensional difference calculation module is divided into three processing engines, corresponding to time domain, frequency domain, and sequence analysis respectively, interconnected via crossbar switches for data distribution. The difference coefficient matrix generation unit is equipped with a high-speed memory interface, supporting real-time construction and transmission of large matrices. The system employs a clock synchronization mechanism to ensure microsecond-level time alignment between theoretical value calculations and measured value acquisition. The data path uses a double-buffered design, pre-fetching the next frame of data while calculating the current frame, eliminating processing latency.

[0032] Example 3: The spatial positioning module of the anomaly analysis unit is responsible for mapping the multidimensional difference analysis results to physical space. This module receives a difference coefficient matrix from the multidimensional difference calculation module, which contains three-dimensional data in terms of time, feature, and difference dimensions. The core of spatial positioning lies in constructing a spatial topology model of the container structure, generating a probability distribution map through anomaly propagation simulation, and ultimately achieving precise positioning of the anomalous physical region.

[0033] The topology modeling submodule first abstracts the physical structure of the composite container into a discretized network model. The container is divided into N regular grid regions, each corresponding to a topological node. Node attributes include static parameters such as geometric center coordinates, material thickness, and sensor distribution density. The connections between nodes are established based on physical connectivity: topological edges are automatically generated between adjacent grid regions, and cross-region connections are established between non-adjacent regions if structural stiffeners or load transfer paths exist. Each topological edge is assigned a weight coefficient.

[0034] in: This represents the connection strength between nodes i and j. For connecting cross-sectional areas (unit: square meters). The elastic modulus of the material (unit: gigapascals). The distance between nodes is the Euclidean distance (in meters). This weighted model reflects the efficiency of mechanical stress transfer and is pre-calculated through finite element analysis and stored in the topology database.

[0035] The anomaly propagation simulation submodule initiates propagation calculations using the difference coefficient matrix as input. During the initialization phase, the difference values ​​of the latest time slice in the matrix are mapped to the corresponding topological nodes, forming an initial anomaly intensity distribution. The propagation process employs an improved diffusion algorithm: within each calculation cycle, the anomaly intensity increment of node i is determined by the intensity gradient of its neighboring nodes.

[0036] in: This represents the anomaly intensity of node i at time t. Let i be the set of adjacent nodes of node i. This is the propagation attenuation factor (default value 0.25). The partial differential equation is solved iteratively using the explicit Euler method, with the time step set to 1 / 10 of the system sampling interval. A strength threshold constraint is set during the iteration process, automatically truncating when the nodal strength exceeds 80% of the material's yield strength. Global normalization is performed after each iteration to prevent exponential growth of the strength value.

[0037] The probability distribution generation submodule transforms the propagated, stable anomaly intensity field into a probability distribution, employing a Bayesian inference framework that treats topological nodes as independent random variables. Prior probabilities are derived from historical anomaly statistics, while the likelihood function is constructed from the current anomaly intensity value. Posterior probability calculation is implemented using Markov chain Monte Carlo sampling, generating a two-dimensional probability matrix. ,in This represents the anomaly probability at grid coordinates (m, n). After smoothing the probability matrix using bicubic spline interpolation, it is converted into a grayscale probability distribution map covering the entire surface of the container, with darker areas corresponding to high-probability anomaly zones.

[0038] Abnormal physical region localization is achieved through probability threshold segmentation, by setting dynamic probability thresholds. ,in The mean probability of the entire graph. The standard deviation is [value]. For probability plots exceeding [value], [the value is] [value]. Connected regions are marked, and the centroid coordinates of each region are calculated and mapped back to the physical coordinate system. The localization results are output as a list of abnormal regions, with each entry containing a region number, center coordinates, boundary polygon vertices, and confidence score. Simultaneously, a visualization engine is triggered to overlay a semi-transparent heatmap onto the 3D container model, with high-probability regions displaying a red gradient effect.

[0039] The entire processing flow is deployed on a graphics processing unit (GPU) platform. Topology model data is stored in texture memory, anomaly propagation calculations are mapped to pixel shader programs, and probabilistic sampling is performed in parallel using computational shaders. The time required for a single spatial localization operation is controlled within 5 milliseconds, and output data is transmitted to the early warning decision unit via the PCIe bus. The system maintains a topology version management mechanism, supporting online model updates when the container structure is modified.

[0040] Example 4: See Figure 4 The adaptive learning module is responsible for dynamically optimizing the state prediction model. This module takes anomaly probability distribution maps as input and adjusts model parameters online through neuron state analysis. Its implementation involves three collaborative sub-modules, forming an adaptive closed loop from neuron identification to learning rate calculation. The activated neuron identification sub-module continuously monitors the intermediate layer output of the dynamic state modeling module. When real-time state feature vectors are input into the prediction model, this sub-module records the activation values ​​of all neurons in the hidden layer. The activation threshold is set to 0.65; when a neuron's output value exceeds this threshold, it is marked as active. The identification process uses a parallel comparator array, completing the screening of 1024 neurons within 200 microseconds. The activated neuron numbers are recorded in a circular buffer, and a binary bitmap is generated to mark the activation state. For example, at a certain moment, the record shows that neuron numbers {15, 28, 73, 209} are active, indicating that these neurons are sensitive to the current input pattern.

[0041] The fuzzy neuron identification submodule analyzes the stability of activated neurons based on historical reference data points. The reference data point set is stored in a flash database, containing records of activated neurons under typical conditions over the past 72 hours. This submodule performs set operations: taking the intersection of the current set of activated neurons and the historical reference set. When the intersection ratio is lower than a preset threshold (e.g., 40%), the current activation pattern is deemed abnormal. At this point, the jump data point analysis unit is activated to calculate the degree of deviation between the current data point and the historical data distribution. The deviation is measured using a multidimensional spatial distance metric, considering the 12 principal components of the feature vector. Data points with a distance value greater than three standard deviations are marked as jump points; for example, a distance value of 7.3 was detected (historical mean 2.1, standard deviation 1.5).

[0042] Based on skipped data points, the fuzzy neuron determination submodule performs weight similarity matching. The feature vectors of the skipped points are extracted, and their cosine similarity is calculated with the weight vectors of all neurons. Neurons with a similarity lower than 0.3 are marked as fuzzy neurons, indicating that their weight configuration does not match the current abnormal pattern. For example, in a certain abnormal event, the similarities of neurons {38, 112, 305} are 0.18, 0.22, and 0.27 respectively, and they are added to the fuzzy neuron list. The system maintains dynamic neuron state records, see Table 1.

[0043] Table 1: Neuron State Record Table

[0044] The learning rate calculation submodule determines the parameter update strategy based on the neuron state. This module first obtains the preset learning rate (default 0.01) for activated neurons, and then calculates the learning rate adjustment coefficient for fuzzy neurons through the position change evaluation unit. The position change evaluation tracks the change trajectory of each neuron's weight vector in the last 20 iterations and calculates the coefficient of variation of the Euclidean distance. Neurons with a coefficient of variation greater than 0.7 are considered unstable, and their learning rate is proportionally amplified. For example, neuron 305 has a coefficient of variation of 0.83, and its learning rate is adjusted to 0.018; neuron 112 has a coefficient of variation of 0.58, and its learning rate remains at 0.01.

[0045] The learning rate calculation process employs a hierarchical weighted mechanism: the learning rate of activated neurons remains at a base value; the learning rate of fuzzy neurons is increased in stages according to the probability of anomalies; and fuzzy neurons with high coefficients of variation receive additional learning rate weights. Finally, a learning rate configuration matrix is ​​generated and written to the parameter register of the dynamic state modeling module via a direct memory access channel. Parameter updates are performed during system idle time slots to avoid interfering with real-time prediction tasks, with each update taking less than 50 milliseconds.

[0046] The entire adaptive learning process employs an event-driven mechanism. The learning process is automatically triggered when a region with a probability value exceeding 0.75 appears in the anomaly probability distribution map. The system records the model prediction error before and after each learning iteration, but this is not used as a basis for adjusting the learning rate; it is only used to monitor the stability of the learning process. Neuron state data is stored in a columnar format, supporting rapid retrieval of historical activation patterns and providing a reference for pattern evolution in subsequent anomaly analysis.

[0047] Example 5: The anomaly decision-making module of the early warning output unit generates an execution strategy based on the anomaly analysis results. This module receives the anomaly probability distribution map from the spatial positioning module and the updated model parameters from the adaptive learning module, and outputs specific operation instructions through multi-level decision logic. Its implementation process integrates a collaborative judgment mechanism of probability analysis, parameter sensitivity assessment, and physical response testing.

[0048] The anomaly decision-making module first analyzes the topological features of the anomaly probability distribution map, extracting three key indicators during the analysis: the maximum probability value and its coordinates, the geometric features of high-probability regions, and the direction of change of the probability gradient. These indicators are automatically identified using image processing algorithms, such as using morphological operations to extract connected regions and edge detection algorithms to determine boundary curvature. Simultaneously, it reads the neuron state records provided by the adaptive learning module, focusing on the distribution density and positional variation amplitude of fuzzy neurons. When fuzzy neurons are concentrated in a specific network layer and the variance of their positional variation exceeds 0.4, they are marked as model structure-sensitive states.

[0049] Based on the above analysis, the decision engine generates a combination of early warning strategies, and the high-frequency monitoring mode activation command is sent to the multimodal sensing unit via the digital control bus. This command includes a sensor channel selection mask and sampling rate configuration parameters, such as increasing the sampling rate of the strain sensor from 1kHz to 2kHz and expanding the bandwidth of the acoustic emission channel from 50kHz to 100kHz. Sampling mode switching employs phase synchronization technology, switching the sampling clock at the signal zero-crossing point to avoid signal truncation. Simultaneously, the data acquisition unit adjusts its buffering strategy, tripling the capacity of the original data buffer to ensure data integrity at high sampling rates.

[0050] The physical disturbance test execution command triggers the external actuators to work collaboratively. The system sends a sweep frequency signal sequence to the piezoelectric vibrator via industrial Ethernet. The signal frequency range is set according to the material characteristics of the abnormal area. Taking a composite laminate as an example, a 20-200Hz linear sweep frequency vibration is applied to the suspected delamination area for 500 milliseconds. Simultaneously, the thermal excitation unit is activated to apply a gradient temperature field not exceeding the material's tolerance limit to the target area. During the test, the multimodal sensing unit enters synchronous acquisition mode, and all sensor channels record the response signal at the highest sampling rate.

[0051] The disturbance test signal is compared with the normal response in the benchmark database. The comparison dimensions include the correlation coefficient of the time-domain waveform, the amplitude deviation of the frequency response function, and the rate of change of the damping characteristics. The analysis results generate an anomaly verification report, which includes a list of deviations of key indicators. When three or more core indicators deviate by more than 15%, the existence of a physical anomaly is confirmed. At this time, the system automatically upgrades the warning level and generates diagnostic information including an anomaly type prediction, such as "Suspected fiber breakage in area B3, confidence level 72%".

[0052] The early warning strategy employs a status feedback mechanism. In high-frequency monitoring mode, the system analyzes the statistical characteristics of incremental data in real time. If the standard deviation continues to decrease, the system automatically restores the normal sampling rate. Physical disturbance tests are configured with safety interruption conditions; the excitation is immediately terminated when the response signal amplitude exceeds a preset safety threshold. All operation records generate event logs containing structured data such as timestamps, execution parameters, and response characteristics, which are transmitted to the remote monitoring center via a secure protocol.

[0053] The main channel sends machine-readable early warning messages via Industrial Internet of Things (IIoT) protocols. The message structure includes fields such as anomaly coordinates, severity level, and recommended actions. The auxiliary channel generates visual alarms, overlaying flashing markers onto a 3D container model. The marker color gradually changes from yellow to red based on probability values. Local audible and visual alarms are activated simultaneously, with alarm frequency positively correlated with the anomaly probability value. The system retains a manual confirmation interface, allowing operators to view diagnostic details and confirm alarm clearance via a human-machine interface. The entire decision-making execution cycle is controlled within 800 milliseconds, ensuring timeliness from anomaly identification to response output.

[0054] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0055] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time status early warning system for composite material containers based on multimodal perception, characterized in that, include: A multimodal sensing unit is used to collect real-time multimodal sensor data from composite material containers. A real-time acceleration processing unit is used to perform matrix multiplication and addition operations on the data collected by the multimodal sensing unit to generate a state feature vector; An anomaly analysis unit is used to perform multidimensional anomaly assessment based on the state feature vector; The early warning output unit is used to generate an early warning signal when the anomaly analysis unit detects an anomaly.

2. The real-time status early warning system for composite material containers based on multimodal perception according to claim 1, characterized in that, The real-time acceleration processing unit includes: Multi-source signal generation module, used to generate multiple parallel signal sources; A weight loading module is used to load weights onto the signals generated by the multi-source signal generation module according to a preset weight vector. The signal modulation module is used to modulate the data collected by the multimodal sensing unit onto the signal after the weight loading module has loaded the weights; The time delay module is used to apply a time delay to the modulated signal; The feature segment extraction module is used to extract feature segments within a specific time window from the delayed signal; The signal superposition module is used to linearly superimpose the extracted feature segments.

3. The real-time status early warning system for composite material containers based on multimodal perception according to claim 2, characterized in that, The real-time acceleration processing unit also includes a dynamic state modeling module; The dynamic state modeling module is used to construct a dynamic state model based on historical state data, and takes the output of the signal superposition module as input to generate theoretical state values.

4. The real-time status early warning system for composite material containers based on multimodal perception according to claim 3, characterized in that, The anomaly analysis unit includes a multidimensional difference calculation module; The multidimensional difference calculation module is used to perform multidimensional difference analysis on the theoretical state values ​​and measured state values ​​generated by the dynamic state modeling module, and generate a difference coefficient matrix. The multidimensional difference analysis includes time-domain cumulative deviation calculation, frequency-domain energy shift detection, and signal sequence similarity evaluation.

5. The real-time status early warning system for composite material containers based on multimodal perception according to claim 4, characterized in that, The anomaly analysis unit also includes a spatial positioning module; The spatial positioning module is used to input the difference coefficient matrix generated by the multidimensional difference calculation module into the spatial topology model, generate an anomaly probability distribution map, and locate the abnormal physical region.

6. The real-time status early warning system for composite material containers based on multimodal perception according to claim 5, characterized in that, The spatial positioning module includes: The topology modeling submodule is used to construct the spatial topology of composite container; An anomaly propagation simulation submodule is used to simulate anomaly propagation paths based on the difference coefficient matrix. The probability distribution generation submodule is used to generate anomaly probability distribution maps covering the entire area.

7. The real-time status early warning system for composite material containers based on multimodal perception according to claim 6, characterized in that, The anomaly analysis unit also includes an adaptive learning module; The adaptive learning module is used to update the model parameters of the dynamic state modeling module based on the anomaly probability distribution map; The adaptive learning module includes: an activation neuron identification submodule, used to identify the activation neurons of the current data point; The fuzzy neuron determination submodule is used to determine fuzzy neurons based on the intersection relationship between reference data points and activated neurons; The learning rate calculation submodule is used to calculate the learning rate of the fuzzy neuron.

8. The real-time status early warning system for composite material containers based on multimodal perception according to claim 7, characterized in that, The fuzzy neuron determination submodule includes a jump data point analysis unit; The jump data point analysis unit is used to calculate the probability of anomalies in data points and determine the target jump data point based on the probability of anomalies. The fuzzy neuron identification submodule is used to identify fuzzy neurons based on target jump data points.

9. The real-time status early warning system for composite material containers based on multimodal perception according to claim 8, characterized in that, The learning rate calculation submodule includes a position change evaluation unit; The position change evaluation unit is used to calculate the variance of neuron position change; The learning rate calculation submodule is used to calculate the learning rate of the fuzzy neuron based on the position variation variance and the preset learning rate of the activated neuron.

10. The real-time status early warning system for composite material containers based on multimodal perception according to claim 9, characterized in that, The early warning output unit includes an anomaly decision-making module; The anomaly decision-making module is used to generate early warning strategies based on the anomaly probability distribution map and the updated model parameters; The early warning strategy includes activating high-frequency monitoring mode and executing physical disturbance tests.

Images