Battery compartment foreign matter identification method and system based on laser scanning and AI identification

Abstract

The application belongs to the technical field of foreign matter detection and recognition, and provides a battery cabin foreign matter recognition method and system based on laser scanning and AI recognition. A laser scanning module is used to scan and collect three-dimensional point cloud data inside the battery cabin at a preset period, the collected original point cloud data is preprocessed and converted into a unified cabin coordinate system, point cloud segmentation is performed, a hypersphere model of normal state point cloud is constructed based on a support vector data description method, an abnormal state is discriminated through a dynamic threshold, foreign matter position tracing is performed according to personnel positioning data combined with multi-period scanning data, and dynamic time warping and space-time filtering are cooperated to improve multi-source data alignment accuracy and trajectory estimation stability. The application can effectively extract complex features, improve foreign matter recognition accuracy, can recognize foreign matter in real time and judge its properties, improve data processing efficiency, can automatically adapt to complex and variable environments, and ensure stable and reliable operation under various working conditions.

Description

Technical Field

[0001] This invention belongs to the field of foreign object detection and identification technology, specifically relating to a method and system for identifying foreign objects in a battery compartment based on laser scanning and AI recognition. Background Technology

[0002] An energy storage power station is a device system that stores, converts, and releases recyclable electrical energy through electrochemical cells or electromagnetic energy storage media. As the core energy storage unit of an energy storage power station, the battery compartment faces risks to its safe operation, such as short circuits and thermal runaway caused by foreign objects inside the battery compartment, such as metal debris, condensate, or animal intrusion.

[0003] In the current field of foreign object identification in the battery compartments of energy storage power stations, the main methods used include traditional manual inspection and simple image recognition technology.

[0004] Manual inspection involves staff periodically entering the battery compartment and using only their eyes and simple tools, such as flashlights and binoculars, to inspect the equipment and surrounding environment. However, traditional manual inspection is limited by human visual acuity. Tiny foreign objects, such as metal fragments smaller than 1 cm or fine fibers, are difficult to detect in the complex environment of the battery compartment. Furthermore, when foreign objects are similar in color or shape to surrounding equipment, misjudgments or omissions are easily made. For example, darker-colored plastic fragments falling onto the black battery casing may be difficult for inspectors to notice.

[0005] Simple image recognition technology: This involves installing cameras inside the battery compartment to capture images of the interior. These images are then transmitted to an image analysis system, which uses basic image processing algorithms, such as edge detection and color recognition, to attempt to identify foreign objects. However, this simple image recognition technology also has limitations. Due to the numerous devices and complex background inside the battery compartment, existing edge detection and color recognition algorithms are easily affected by noise when processing images, leading to inaccurate extraction of the edges and color features of foreign objects. This results in an inability to accurately distinguish between small foreign objects or similar objects, increasing the probability of false positives and false negatives.

[0006] As energy storage power stations expand in scale and the number of battery compartments increases, the amount of data requiring monitoring grows exponentially. Manual inspections are time-consuming and labor-intensive, with long inspection cycles, making real-time monitoring impossible. For example, a large energy storage power station may contain dozens of battery compartments; a manual inspection could take hours or even days. If foreign objects enter and cause safety issues, timely detection and handling are difficult. Simple image recognition technologies, due to their relatively simple image analysis algorithms, are slow to process large amounts of image data. When real-time analysis of continuous video images is required, existing processing units may not be able to complete the analysis of all images within a short time, leading to monitoring delays and failing to meet the demands of real-time monitoring.

[0007] The battery compartment has a complex environment with varying lighting conditions and equipment obstructions. The intensity and angle of light within the compartment change at different times of day, impacting both manual inspection and image recognition. For example, in the dim light of early morning and late afternoon, manual inspections are prone to missing foreign objects, while image recognition systems may suffer from reduced image quality due to insufficient light, failing to accurately identify foreign objects. Furthermore, the battery compartment contains densely packed equipment and numerous obstructions such as battery racks and cable trays. During manual inspections, some obstructed areas may not be fully inspected, and image recognition technology, lacking complete information, struggles to accurately determine the presence of foreign objects when processing images of obstructed areas, severely affecting recognition performance. Summary of the Invention

[0008] The purpose of this invention is to overcome the existing defects and provide a method and system for identifying foreign objects in a battery compartment based on laser scanning and AI recognition.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0010] A method for identifying foreign objects in a battery compartment based on laser scanning and AI recognition includes:

[0011] The laser scanning module is used to scan and collect three-dimensional point cloud data inside the battery compartment at a preset cycle.

[0012] The collected raw point cloud data is preprocessed and converted into a unified cabin coordinate system;

[0013] An improved PointNet++ architecture is used for point cloud segmentation. A hypersphere model of the normal state point cloud is constructed based on the support vector data description method, and abnormal states are distinguished by dynamic threshold.

[0014] By acquiring personnel location data from the access control system and combining it with multi-time period scanning data, foreign object location can be traced. Dynamic time warping and spatiotemporal filtering are used to improve the alignment accuracy of multi-source data and the stability of trajectory estimation.

[0015] Furthermore, in the process of using the laser scanning module to scan and collect 3D point cloud data inside the battery compartment at a preset cycle, the scanning path of the laser scanning module is planned based on an improved Hilbert curve model. In dense areas of the battery rack, the path density is dynamically adjusted according to the local complexity of the point cloud reconstruction model. During the scanning process, the resolution is adaptively adjusted according to the accuracy requirements of the detection task.

[0016] Furthermore, the step of preprocessing the collected raw point cloud data and converting it into a unified cabin coordinate system includes:

[0017] Dynamic background modeling: Based on the adaptive Gaussian mixture model, the fixed structure and static equipment in the battery compartment are modeled. The point cloud information that has existed for a long time and has a stable shape is extracted as the baseline background. The probability distribution of each point is continuously updated to eliminate false differences caused by changes in lighting or slight vibrations.

[0018] Outlier filtering: A statistical filtering algorithm is used to identify and remove spatially isolated points. The average distance from each point to its surrounding points in its local neighborhood is calculated and compared with the overall distribution. Points whose average distance from their neighborhood exceeds a preset threshold are filtered out.

[0019] Data augmentation: Introducing small perturbations into the original point cloud to improve the model's adaptability to changes in pose and position;

[0020] UWB positioning data processing: A density-based outlier detection method is introduced, and a multi-scale judgment strategy is constructed by integrating a statistical filtering mechanism to handle the interpolation error caused by sudden offset points in the UWB trajectory;

[0021] Coordinate system transformation: Based on the pre-calibrated cabin reference points, a global coordinate mapping relationship is established. Spatial alignment of point cloud data is achieved through rotation and translation transformation matrices, and point cloud data collected at different times and angles are fused under the same spatial reference.

[0022] Furthermore, the improved PointNet++ architecture is used for point cloud segmentation, a hypersphere model of the normal state point cloud is constructed based on the support vector data description method, and abnormal states are identified through dynamic thresholds, including:

[0023] Point cloud segmentation: An improved PointNet++ architecture is adopted, and deformable convolutional layers are introduced into the encoder to adapt to the non-rigid and local deformation information in the cabin structure. At the same time, a multi-scale feature extraction path is constructed to capture spatial details at different scales.

[0024] Foreign object classification: Based on the support vector data description method, a hypersphere model of the normal state point cloud is constructed. Normal data is tightly surrounded in the feature space, and abnormal states are judged by dynamic threshold. Points outside the threshold are marked as potential foreign objects.

[0025] Confidence fusion: Stability analysis of detection results is performed in the time series dimension. Kalman filtering is used to smooth the output probability distribution of N consecutive frames, and the confidence of the presence of foreign objects in each frame is calculated. When the cumulative abnormal probability over time exceeds a set value, an alarm mechanism is automatically triggered.

[0026] Furthermore, the improved PointNet++ architecture includes:

[0027] Deformable convolutional layers are introduced into the encoder to adapt to the non-rigid and local deformation information in the cabin structure. At the same time, a multi-scale feature extraction path is constructed, and 1×1×1, 3×3×3, and 5×5×5 convolutional kernels are used in the parallel branch to capture spatial details at different scales.

[0028] The decoder stage introduces a dual attention mechanism of spatial and channel attention, dynamically adjusting the response intensity of each dimension in the feature map to weight and enhance the loss function. The optimized loss function is expressed as follows:

[0029]

[0030] Where S is the predicted point cloud output by the model, For real point clouds;

[0031] x represents any point in the predicted point cloud S;

[0032] y represents the true point cloud any point in it;

[0033] ||xy|| 2 This represents the squared Euclidean distance between point x and point y, used to quantify the positional error between the two points.

[0034] This represents the squared distance from the predicted point x to the nearest point y in the true point cloud;

[0035] This represents the squared distance from the true point y to the nearest point x in the predicted point cloud.

[0036] Furthermore, the acquisition of personnel positioning data from the access control system, combined with multi-time period scanning data for foreign object location tracing, and the improvement of multi-source data alignment accuracy and trajectory estimation stability through dynamic time warping and spatiotemporal filtering, include:

[0037] Multi-source data acquisition: Acquire personnel location data from the access control system, as well as scanning data from multiple time periods;

[0038] Dynamic time warping: A dynamic time warping method based on an event triggering mechanism is adopted. Scanning nodes are triggered when personnel enter or exit a specific area, so that the trajectory time series and the laser scanning time series are non-linearly aligned on the event point set. The trajectory node alignment process is constrained by minimizing the time difference cost function.

[0039] Probability density modeling: Based on the spatial interpolation algorithm unit, Kriging interpolation is introduced to perform probability density modeling on discrete point clouds, and an estimation model is constructed for the spatial location of the point cloud;

[0040] Joint Kalman Filtering: For the fusion of sensors with different accuracies, the observation model combines laser point cloud and UWB positioning information, sets point cloud error and UWB error, and performs weighted processing by assigning different observation error covariances. The process noise adopts an adaptive covariance matrix to reflect the system uncertainty in positioning and motion prediction, and dynamically adjusts the filtering response capability.

[0041] Furthermore, in the multi-source data acquisition, a timestamp mapping function is established using a timestamp calibration method, and the historical data used includes personnel trajectory data and laser scan data from at least the past week.

[0042] Another objective of this invention is to provide a foreign object identification system for a battery compartment based on laser scanning and AI recognition, characterized in that the system comprises:

[0043] The laser scanning module consists of a linear array lidar mounted on a programmable rotating bracket, used to collect three-dimensional point cloud data of the battery compartment interior at preset intervals.

[0044] The data processing module, connected to the laser scanning module, is used to denoise the raw point cloud data and convert it into a unified cabin coordinate system.

[0045] The AI ​​recognition module, connected to the data processing module, contains a trained foreign object detection neural network. It is used to build a benchmark model based on historical normal state point cloud datasets and identify foreign objects by comparing real-time point cloud differences.

[0046] The location tracking module is used to acquire personnel location data from the access control system and combine it with multi-time period scanning data to trace the location of foreign objects.

[0047] The work order interface module, connected to the AI ​​recognition module and the location tracking module, is used to generate maintenance work orders that include the coordinates of foreign objects and trajectory traceability information.

[0048] Furthermore, the data processing module includes: a point cloud preprocessing unit, used for dynamic background modeling, outlier removal from laser scanning data, and processing of abnormal interference in UWB positioning data;

[0049] The coordinate system transformation unit is used to convert the raw point cloud data into a unified cabin coordinate system.

[0050] Furthermore, the AI ​​recognition module includes:

[0051] The point cloud segmentation unit adopts an improved PointNet++ architecture, introduces deformable convolutional layers in the encoder to adapt to the non-rigid and local deformation information in the cabin structure, and constructs a multi-scale feature extraction path to capture spatial details at different scales.

[0052] The foreign object classification unit constructs a hypersphere model of the normal state point cloud based on the support vector data description method. It tightly surrounds the normal data in the feature space and uses a dynamic threshold to distinguish abnormal states. Points outside the threshold are marked as potential foreign objects.

[0053] The confidence fusion unit is used to perform stability analysis of detection results in the time series dimension. It uses Kalman filtering to smooth the output probability distribution of N consecutive frames, calculates the confidence of the presence of foreign objects in each frame, and automatically triggers an alarm mechanism when the cumulative abnormal probability over time exceeds a set value.

[0054] In combination with the above technical solutions, the beneficial effects of the present invention compared with the prior art are as follows:

[0055] This invention uses a laser scanning module to acquire high-resolution 3D point cloud data, accurately capturing fine structures and object information. At the same time, the AI ​​recognition module utilizes advanced algorithms such as the improved PointNet++ architecture to more effectively extract complex features, thereby improving the accuracy of foreign object recognition.

[0056] This invention leverages the powerful data analysis and learning capabilities of AI algorithms to rapidly process and analyze large amounts of point cloud data acquired through laser scanning. It can identify foreign objects and determine their properties in real time, significantly improving data processing efficiency.

[0057] This invention optimizes algorithms and system design, such as the dynamic background modeling unit to eliminate false differences due to lighting variations, the AI ​​recognition module to adapt to complex geometries and environments using various optimization methods, and the position tracking module to achieve multi-source data fusion and precise spatiotemporal positioning through various technologies. This enables the system to automatically adapt to complex and ever-changing environments and ensures stable and reliable operation under various working conditions.

[0058] This invention constructs a complete system architecture that includes laser scanning, data processing, AI recognition, location tracking, and work order interface. The modules work together to achieve closed-loop management of the foreign object identification process. The location tracking module uses multi-source heterogeneous data fusion technology to accurately trace the location of foreign objects, improve the accuracy of judging the intersection of abnormal areas and personnel behavior, and generate maintenance work orders containing foreign object coordinates and trajectory tracing information for convenient subsequent processing. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a flowchart of a foreign object identification method for a battery compartment based on laser scanning and AI recognition provided in an embodiment of the present invention;

[0061] Figure 2 This is a structural schematic diagram of the foreign object identification system for the battery compartment based on laser scanning and AI recognition provided in an embodiment of the present invention;

[0062] Figure 3 This is a structural schematic diagram of the data processing module provided in an embodiment of the present invention;

[0063] Figure 4 and Figure 5 This is a system signaling diagram of a foreign object identification system for a battery compartment based on laser scanning and AI recognition, provided in an embodiment of the present invention. Detailed Implementation

[0064] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0065] Example 1:

[0066] like Figure 1 The image shows an embodiment of the foreign object identification method and system for battery compartments based on laser scanning and AI recognition provided by the present invention, including:

[0067] S1: Use a laser scanning module to scan and collect three-dimensional point cloud data inside the battery compartment at a preset cycle;

[0068] S2: Perform point cloud preprocessing on the collected raw point cloud data and convert it into a unified cabin coordinate system;

[0069] S3: Point cloud segmentation is performed using an improved PointNet++ architecture. A hypersphere model of the normal state point cloud is constructed based on the support vector data description method, and abnormal states are identified by dynamic thresholds.

[0070] S4: Acquire personnel positioning data from the access control system, combine multi-time period scanning data to trace the location of foreign objects, and improve the alignment accuracy of multi-source data and the stability of trajectory estimation through dynamic time warping and spatiotemporal filtering.

[0071] Preferably, in step S1, the installation spacing of the laser scanning modules is designed to ensure complete coverage of the cabin space and minimize blind spots, and is configured using geometric relationships. Specifically, the linear array lidar is deployed inside the cabin via a programmable rotating bracket, and the installation spacing satisfies the formula...

[0072]

[0073] Where D represents the minimum horizontal installation distance between adjacent lidars, H is the vertical height inside the battery compartment, and θ is the vertical field of view angle of the lidar. This formula is derived based on the spatial projection relationship between the radar's conical field of view and the compartment height, ensuring that the scanning sectors of adjacent devices intersect or slightly overlap in the horizontal plane to eliminate data loss caused by radar blind spots. The scanning path planning is based on an improved Hilbert curve model, which has good space-filling performance and path continuity, effectively improving the coverage efficiency of complex structural surfaces inside the compartment. In dense battery rack areas, the path density is dynamically adjusted according to the local complexity of the point cloud reconstruction model, realizing a high-frequency retracement strategy to improve the point cloud sampling accuracy in key areas. During the scanning process, the system adaptively adjusts the resolution according to the accuracy requirements of the detection task. In the basic state, a standard point cloud model is generated by sampling at equal intervals of 5 mm. Once the detection algorithm identifies a suspected foreign object or abnormal structure, the scanning module immediately locally increases the resolution to 2 mm, enhancing the detail capture capability through dense point cloud sampling, thereby providing higher quality spatial data support for subsequent point cloud processing and foreign object identification.

[0074] It should be noted that the improved Hilbert curve pattern planning is based on the traditional Hilbert curve space-filling characteristics and is optimized for the complex structural distribution within the battery compartment. Specifically, it involves: analyzing the local complexity of the point cloud reconstruction model, dynamically increasing the scanning path density in densely packed battery rack areas, and employing a high-frequency retrace strategy to improve the point cloud sampling accuracy in key areas; simultaneously, adaptively adjusting the resolution during scanning based on the accuracy requirements of the detection task. In the basic state, a standard point cloud model is generated using 5mm equal-interval sampling. When a suspected foreign object or abnormal structure is detected, the local resolution is automatically increased to 2mm, enhancing the ability to capture details through dense point cloud sampling. Applying this improved method effectively improves the coverage efficiency of complex structural surfaces inside the compartment, eliminates data loss caused by radar blind spots, provides higher-quality spatial data support for subsequent point cloud processing and foreign object identification, and significantly improves the detection accuracy of small foreign objects and adaptability in complex environments.

[0075] Preferably, in step S2, the point cloud preprocessing of the collected raw point cloud data and its conversion into a unified cabin coordinate system includes:

[0076] Dynamic background modeling: Based on the adaptive Gaussian mixture model, the fixed structure and static equipment in the battery compartment are modeled. The point cloud information that has existed for a long time and has a stable shape is extracted as the baseline background. The probability distribution of each point is continuously updated to eliminate false differences caused by changes in lighting or slight vibrations.

[0077] Outlier filtering: A statistical filtering algorithm is used to identify and remove spatially isolated points. The average distance from each point to its surrounding points in its local neighborhood is calculated and compared with the overall distribution. Points whose average distance from their neighborhood exceeds a preset threshold are filtered out.

[0078] Data augmentation: Introducing small perturbations into the original point cloud to improve the model's adaptability to changes in pose and position;

[0079] UWB positioning data processing: A density-based outlier detection method is introduced, and a multi-scale judgment strategy is constructed by integrating a statistical filtering mechanism to handle the interpolation error caused by sudden offset points in the UWB trajectory;

[0080] Coordinate system transformation: Based on the pre-calibrated cabin reference points, a global coordinate mapping relationship is established. Spatial alignment of point cloud data is achieved through rotation and translation transformation matrices, and point cloud data collected at different times and angles are fused under the same spatial reference.

[0081] Specifically, point cloud preprocessing of the collected raw point cloud data aims to improve the robustness and accuracy of the subsequent recognition model to the input data. First, the dynamic background modeling unit models the fixed structures and static equipment within the battery compartment based on an adaptive Gaussian mixture model. Long-term existing and morphologically stable point cloud information is extracted as the baseline background for accurate detection of subsequent dynamic changes. This model continuously updates the probability distribution of each point, gradually eliminating false differences caused by changes in illumination or minor vibrations. Second, the outlier filtering unit uses a statistical filtering algorithm to identify and remove spatially isolated points. Within each point's local neighborhood, the average distance to surrounding points is calculated and compared with the overall distribution. The set filtering parameters include a neighborhood radius r = 50 mm and a standard deviation factor k = 2.0. The filtering condition is that the average distance between a point and its neighborhood exceeds the global mean plus k times the standard deviation.

[0082]

[0083] in Let be the average neighborhood distance of the i-th point, and μ and σ be the average distance and standard deviation of all points, respectively. This strategy effectively removes discrete points caused by laser interference or reflection. Finally, the data augmentation unit improves the model's adaptability to pose and position changes by introducing small perturbations on the original point cloud. Augmentation methods include random rotation ±2° and random translation ±5mm, which expand the number of training samples while ensuring that the spatial structure remains basically unchanged, thereby improving the generalization ability of the neural network.

[0084] Point cloud preprocessing not only performs refined outlier removal on laser scanning data but also extends to handling anomalous interference in UWB positioning data. To address interpolation errors caused by sudden shifts in the UWB trajectory, a density-based outlier detection method is introduced, and a multi-scale judgment strategy is constructed by integrating a statistical filtering mechanism. During the processing of each UWB positioning point, the system analyzes the distribution of its neighboring points within a unit time window, calculates the local density index, and determines whether it is consistent with the surrounding trajectory structure. Density anomalies are considered interference and filtered out. Density judgment is based on the Local Outlier Factor (LOF) calculation. The LOF value of each point is defined as the ratio of its reachable density to the average reachable density of its neighborhood. If the LOF value satisfies the condition...

[0085] LOF i >δ

[0086] Among them LOF i Let δ be the local outlier factor for the i-th UWB location point, and δ be a set threshold. The system determines this point as an outlier. Based on this, a statistical filtering strategy is further superimposed to construct a dynamic window on the time series. Within a sliding interval of length TTT, the mean μ and standard deviation σ of the changes in the position of the trajectory points are statistically analyzed, and points satisfying |x... i Outliers where -μ|>βσ|, where x i Here, represents the current location coordinates, and β is the sensitivity coefficient. This dual filtering mechanism effectively eliminates positioning anomalies caused by momentary occlusion, signal reflection, or equipment interference, ensuring the accuracy and robustness of subsequent interpolation, alignment, and matching calculations based on trajectory data in terms of continuity and stability.

[0087] The coordinate transformation step is implemented through a coordinate system transformation unit. Its core function is to convert the raw point cloud data collected by the laser scanning module from the device coordinate system to a unified cabin coordinate system, thereby eliminating errors caused by differences in the spatial positions of different sensors. This process establishes a global coordinate mapping relationship based on pre-calibrated cabin reference points. Spatial alignment of the point cloud data is achieved through rotation and translation transformation matrices, ensuring that point cloud data collected at different times and angles are fused under the same spatial reference. This provides a unified spatial reference framework for subsequent point cloud segmentation, foreign object identification, and location tracing, improving the spatiotemporal consistency and analysis accuracy of multi-source data.

[0088] Preferably, step S3, which uses an improved PointNet++ architecture for point cloud segmentation, specifically includes:

[0089] The point cloud segmentation network adopts an improved PointNet++ architecture. Deformable convolutional layers are introduced in the encoder to adapt to the non-rigid and local deformation information present in the cabin structure. Simultaneously, a multi-scale feature extraction pathway is constructed, using 1×1×1, 3×3×3, and 5×5×5 convolutional kernels in the parallel branches to capture spatial details at different scales, effectively improving the ability to represent complex geometric shapes. The decoder stage introduces a dual spatial and channel attention mechanism to dynamically adjust the response intensity of each dimension in the feature map, achieving significant enhancement of key regions. In the loss function design, the FocalLoss approach is used to weight and strengthen the traditional Chamfer Distance loss function. The optimized loss function is expressed as follows:

[0090]

[0091] Where S is the predicted point cloud output by the model, Let x represent any point in the predicted point cloud S, and y represent the true point cloud. Any point in ||xy|| 2 This represents the squared Euclidean distance between point x and point y, used to quantify the positional error between the two points. This represents the squared distance from the predicted point x to the nearest point y in the true point cloud; This represents the squared distance from the ground truth point y to the nearest point x in the predicted point cloud. This function improves sensitivity to sparse and foreign structures while maintaining the geometric alignment accuracy of the point cloud.

[0092] Preferably, a hypersphere model of the normal state point cloud is constructed based on the support vector data description method, and abnormal states are identified through dynamic thresholds, including:

[0093] A hypersphere model of normal-state point clouds is constructed based on the support vector data description method. Normal data is tightly enclosed in the feature space, and a dynamic threshold α = μ + 3σ is used to achieve high-precision discrimination of abnormal states, where μ is the average distance from normal sample points to the center of the hypersphere, and σ is its standard deviation. Points outside the threshold are marked as potential foreign objects. A confidence fusion unit performs stability analysis of the detection results in the time series dimension, using Kalman filtering to smooth the output probability distribution of N consecutive frames, and calculating the confidence level p for the presence of foreign objects in each frame. i The cumulative anomaly probability over time is P = 1 - Π(1 - p). i When the value is greater than 0.95, an alarm mechanism is automatically triggered to suppress momentary misjudgments and improve system stability by fusing time-domain information.

[0094] Preferably, in step S4, personnel positioning data from the access control system is acquired, and foreign object location is traced by combining multi-time period scanning data. Dynamic time warping and spatiotemporal filtering are used to collaboratively improve the alignment accuracy of multi-source data and the stability of trajectory estimation, including:

[0095] Time synchronization and spatial positioning fusion: The system introduces an event-triggered dynamic time warping sub-unit to enhance time registration accuracy by leveraging key behavioral nodes, building upon the traditional sliding window dynamic time warping algorithm. When events with clear markers, such as personnel entering or leaving a specific area, occur, the system immediately initiates a forced synchronization strategy, breaking the passive sliding mode under a fixed time step. This achieves precise alignment of trajectory data and laser scan data at the event point, significantly reducing timing lag errors caused by delays. In scenarios where multi-source data sampling frequencies differ (trajectory data is recorded at 1Hz, while laser scan data is sampled at 0.2Hz), a sliding window mechanism is used for time synchronization. The window length for each processing step is set to 5 seconds, sliding along the time axis in 1-second increments. Within each window, an optimal matching path is constructed, ensuring that data point pairs with inconsistent sampling intervals establish a minimum-cost matching path in the time domain. The cost function is based on the time difference. Based on, among which This represents the timestamp of the i-th trajectory data point, in seconds, corresponding to the location data recorded by UWB or other positioning systems at the i-th time. This represents the timestamp of the j-th laser scan data frame, in seconds, corresponding to a point cloud acquisition completed by the lidar at the j-th time; Δt ij This represents the absolute difference in time between the i-th trajectory point and the j-th scan frame, used for calculating the time cost of matching trajectory points and scan frames during dynamic time warping. Dynamic planning of the alignment path is achieved under the objective of minimizing the total time difference, ensuring that data at the event trigger point can stably fall within the time window corresponding to adjacent scan frames, thus improving the fusion accuracy of the trajectory and scan.

[0096] The timestamp calibration subunit addresses the inconsistency in acquisition frequency and timestamp accuracy between laser scan data and personnel trajectory data. It constructs a time mapping function based on regression learning and trains a prediction model using historical trajectory data to accurately estimate locations not captured at the scan time. Assuming the UWB positioning system generates trajectory points at a frequency of 1Hz, this subunit extracts time-series information from continuous trajectory segments to construct a functional relationship between time and position. It then trains a multivariate regression model using supervised learning to perform the calculation for a given scan time point t. s Predict the location of the corresponding personnel The mapping function is defined as follows

[0097]

[0098] Where (t) i ,P i () represents the i-th moment in the historical trajectory and its corresponding position. Let t be the scan time. s The corresponding predicted location, The training sample sequence provided for the model, where t i P is the timestamp of the i-th historical moment. i This represents the actual location of the person at that moment, and n is the number of trajectory points used for modeling. These data are used together to train the function f, enabling it to capture the time-series characteristics of person movement. By selecting regression models with good generalization ability, such as support vector regression, LSTM regression networks, or multinomial regression, for training, the system improves the accuracy of time registration without introducing hard synchronization from multiple sources, ensuring the synchronization and consistency of the prediction results with the laser scanning point cloud data in the spatial dimension, thereby enhancing the accuracy of trajectory-environment information matching.

[0099] The hierarchical Kalman filter subunit achieves systematic suppression of UWB positioning errors and high-precision modeling of continuous trajectories by constructing a multi-layered filtering architecture. In the first layer, the high spatial resolution of the laser point cloud data is used to correct high-frequency noise in the original UWB trajectory data. The spatial accuracy of the lidar in scanning the cabin structure reaches ±3 mm. This accuracy is used as the measurement update input to the Kalman filter, effectively enhancing the confidence of position observation and enabling smooth correction of UWB measurement points under local environmental constraints. A state vector is introduced in the second layer.

[0100]

[0101] Describe the position and velocity of the person in three-dimensional space using state transition equations.

[0102] x k+1 =F k x k +w k

[0103] Simulate the motion evolution of personnel at continuous time steps, where F k Let w be the state transition matrix. k This represents process noise, reflecting the dynamic uncertainty of the system. Observation updates employ a joint measurement model.

[0104] z k =H k x k +v k

[0105] Where z k For the fused multi-source measurements, H kLet v be the observation matrix. k To mitigate noise, in the prediction-update filtering loop, the system dynamic model is used to estimate the personnel's position status at the next moment, while the measurement model incorporates point cloud correction data and UWB positioning values ​​for error feedback adjustment. By jointly optimizing spatial position and velocity information, continuous modeling of discrete trajectory points is achieved, significantly reducing positioning jumps and drift problems caused by low-precision sensors. The resulting smooth trajectory provides a stable and reliable data foundation for foreign object detection and trajectory tracing in the spatiotemporal dimensions.

[0106] The adaptive weighted fusion subunit enhances the intelligence of multi-source data fusion by introducing an environmental perception mechanism. The system adjusts the fusion weights of laser point cloud data and UWB positioning data in real time based on the complexity of the battery compartment's internal structure. For spatially heterogeneous distributions, a Bayesian inference model is used to construct a posterior probability distribution for the position estimate at each time step.

[0107] P(P t |Z t )∝P(Z t |P t )P(P t )

[0108] Where P(P) t |Z t Let P(Z) be the posterior probability of the fused location information. t |P t P(P) represents the likelihood of the observed value Zt given at the current position Pt. t The prior location distribution is represented by the laser point cloud. In areas with dense battery racks, the boundary contours and obstacle distributions provided by the laser point cloud have high confidence, and the system automatically increases their weight by dynamically adjusting the parameter λ. lidar ,λ uwb A weighted formula is introduced in the fusion location estimation.

[0109]

[0110] Where λ lidar +λ uwb =1, For the estimation of the merged personnel locations, P t lidar The location is estimated based on laser point cloud data, P t uwb To achieve optimal integration of heterogeneous information for location estimation based on UWB positioning data, the system simultaneously generates a probability heatmap in 3D space, representing the probability of a person's spatial presence at different times, serving as input support for subsequent interpolation optimization.

[0111] The spatiotemporal database enhancement subunit supplements the original trajectory database with environmental state variables corresponding to the scan time, defining triples (T... i ,P i E i Let , i, j ...

[0112] The inverse interpolation optimization subunit constructs a spatial probability density field based on the Kriging interpolation method, and performs this on the observation point set. Constructing the estimation function

[0113]

[0114] in Let w be the probability density of the location of the point to be estimated. i The weights are calculated from the spatial covariance matrix, f(x) i At a given point x i The system uses the observed values ​​or location probability density at a given location, where n is the number of known points used for interpolation. For multi-time-segment trajectory and scan data, the system employs dynamic time warping to align the trajectory sequences, extracting potentially related spatiotemporal segments. Based on the estimated probability density function, it generates multiple sets of candidate locations using Monte Carlo simulation, forming the final location search area. Where P(p j Let R be the set of possible locations where the remnant can appear, and p > ∈ R. j For the j-th candidate location point, P(p j ) represents the candidate position p j The probability of occurrence is ∈, which is a set probability threshold. Only candidate points higher than this value are retained to form a search area. m is the number of candidate location points that meet the conditions. This effectively compresses the search space and improves retrieval efficiency and the operability of locating residual foreign objects.

[0115] The dynamic time warping unit addresses the inconsistency in sampling frequency between laser scanning and personnel trajectory. It employs an event-triggered dynamic time warping method, triggering scanning nodes when personnel enter or exit specific areas, thus ensuring the trajectory time series T... uwb ={t1,t2,...,t n} and laser scanning time series T scan ={s1,s2,...,s m Non-linear alignment is achieved on the event point set {ek} by minimizing the time difference cost function. Constrained trajectory node alignment process, Let s be the timestamp of the person's trajectory corresponding to the k-th event point.jk Let be the laser scanning timestamp corresponding to the k-th event point, and D be the sum of the overall nonlinear alignment errors between time series, used to measure the degree of synchronization. The spatial interpolation algorithm unit introduces Kriging interpolation to model the probability density of the discrete point cloud, and constructs an estimation formula for the point cloud spatial location x0:

[0116]

[0117] Where z(x) i Let x be the i-th known spatial location. i The observed value, λ i The optimal interpolation weights, calculated based on the covariance function, satisfy the unbiasedness and minimum variance constraints. The joint Kalman filter section then introduces motion state modeling, assuming the system state vector is X = [x, y, z, v]. x ,v y ,v z The system state transition equation is X. k+1 =AX k +w k The observation model is Z. k =HX k +v k Where A is the state transition matrix, H is the observation matrix, and wk and vk are the process noise and observation noise, respectively. For sensor fusion with different precision levels, the observation model combines laser point cloud and UWB positioning information. The point cloud error is set to ±3mm, and the UWB error to ±100mm. The system performs weighted processing by assigning different observation error covariances. An adaptive covariance matrix is ​​used for the process noise.

[0118]

[0119] Where σ x =σ y =σ z =10mm is the standard deviation of process noise in each axis of the position state, reflecting the uncertainty of position prediction, σ v The standard deviation of process noise at a velocity of 5 mm / s along each axis reflects the perturbation range of the velocity estimate and dynamically adjusts the filtering response capability. This filtering structure, as the second layer of the hierarchical Kalman filter, forms a cascade with the point cloud correction results of the first layer, ultimately outputting a high-confidence continuous trajectory estimate to support reverse reasoning and localization tracing of foreign object locations.

[0120] The timestamp mapping function established by the timestamp calibration subunit of the location tracking module uses historical data including personnel trajectory data and laser scan data from at least the past week.

[0121] Example 2:

[0122] like Figure 2 As shown, this embodiment of the invention provides a foreign object identification system for a battery compartment based on laser scanning and AI recognition, comprising:

[0123] The laser scanning module consists of a linear array lidar mounted on a programmable rotating bracket, used to collect three-dimensional point cloud data of the battery compartment interior at preset intervals.

[0124] The data processing module, connected to the laser scanning module, is used to denoise the raw point cloud data and convert it into a unified cabin coordinate system.

[0125] The AI ​​recognition module, connected to the data processing module, contains a trained foreign object detection neural network. It is used to build a benchmark model based on historical normal state point cloud datasets and identify foreign objects by comparing real-time point cloud differences.

[0126] The location tracking module is used to acquire personnel location data from the access control system and combine it with multi-time period scanning data to trace the location of foreign objects.

[0127] The work order interface module, connected to the AI ​​recognition module and the location tracking module, is used to generate maintenance work orders that include the coordinates of foreign objects and trajectory traceability information.

[0128] Preferably, the data processing module in this embodiment of the invention includes: a point cloud preprocessing unit, used for dynamic background modeling, removing outliers from laser scanning data, and processing abnormal interference in UWB positioning data;

[0129] The coordinate system transformation unit is used to convert the raw point cloud data into a unified cabin coordinate system.

[0130] Preferably, the AI ​​recognition module in this embodiment of the invention includes:

[0131] The point cloud segmentation unit adopts an improved PointNet++ architecture, introduces deformable convolutional layers in the encoder to adapt to the non-rigid and local deformation information in the cabin structure, and constructs a multi-scale feature extraction path to capture spatial details at different scales.

[0132] The foreign object classification unit constructs a hypersphere model of the normal state point cloud based on the support vector data description method. It tightly surrounds the normal data in the feature space and uses a dynamic threshold to distinguish abnormal states. Points outside the threshold are marked as potential foreign objects.

[0133] The confidence fusion unit is used to perform stability analysis of detection results in the time series dimension. It uses Kalman filtering to smooth the output probability distribution of N consecutive frames, calculates the confidence of the presence of foreign objects in each frame, and automatically triggers an alarm mechanism when the cumulative abnormal probability over time exceeds a set value.

[0134] In this embodiment of the invention, the various modules are interconnected through a reasonable connection method and collaborative working mechanism, forming an organic whole. The laser scanning module is responsible for acquiring high-precision 3D point cloud data, providing a data foundation for subsequent analysis; the data processing module preprocesses and transforms the coordinate system of the raw point cloud data, enabling it to be effectively processed by the AI ​​recognition module; the AI ​​recognition module uses advanced neural network algorithms to accurately identify foreign objects; the location tracking module achieves precise tracking of the location of foreign objects through multi-source heterogeneous data fusion; and the work order interface module transforms the identification and tracking results into operable maintenance work orders, realizing closed-loop management of the entire foreign object identification process. This system architecture design fully leverages the advantages of each module, improves the overall performance and reliability of the system, and has significant differences and advantages compared with existing technologies.

[0135] Key Algorithms: In the AI ​​recognition module, the point cloud segmentation network adopts an improved PointNet++ architecture, particularly by adding deformable convolutional layers in the encoder section, and introducing multi-scale feature extraction structures and spatial-channel dual attention weights. Compared with traditional point cloud segmentation algorithms, this can more effectively extract complex features from point cloud data, improving the accuracy and efficiency of foreign object recognition. The foreign object classification model is based on constructing a normal-state hypersphere using Support Vector Data Description (SVDD) and setting dynamic thresholds. Compared with traditional classification algorithms, it can better adapt to the complex environment and data distribution within the battery compartment of energy storage power stations, accurately classifying and judging foreign objects. In the location tracking module, the dynamic time warping unit uses an event-triggered dynamic time warping sub-unit method, and the spatiotemporal joint Kalman filter algorithm can effectively solve the problem of time synchronization and spatial positioning accuracy fusion when processing multi-source heterogeneous data, improving the accuracy of foreign object location tracing.

[0136] Data processing flow: Starting with laser scanning data acquisition, the data processing module performs noise reduction and coordinate system transformation, followed by foreign object identification by the AI ​​recognition module, location tracking by the location tracking module, and finally, the work order interface module generates a maintenance work order. This entire data processing flow is unique and advantageous. Through a series of preprocessing, feature extraction, model training, and data fusion operations, it achieves efficient transformation from raw data to the final foreign object identification and processing results. This data processing flow not only improves the system's performance and accuracy but also has good scalability and adaptability, meeting the foreign object identification needs of battery compartments in energy storage power stations of varying sizes and complexities. This is the key difference between this invention and existing technologies and should be protected by patent.

[0137] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated in the present invention, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0138] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0139] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying foreign objects in a battery compartment based on laser scanning and AI recognition, characterized in that, The method includes: The laser scanning module is used to scan and collect three-dimensional point cloud data inside the battery compartment at a preset cycle. The collected raw point cloud data is preprocessed and converted into a unified cabin coordinate system; An improved PointNet++ architecture is used for point cloud segmentation. A hypersphere model of the normal state point cloud is constructed based on the support vector data description method, and abnormal states are distinguished by dynamic threshold. Acquire personnel positioning data from the access control system, combine multi-time period scanning data to trace the location of foreign objects, and improve the alignment accuracy of multi-source data and the stability of trajectory estimation through dynamic time warping and spatiotemporal filtering. The process of using an improved PointNet++ architecture for point cloud segmentation, constructing a hypersphere model of the normal-state point cloud based on the support vector data description method, and identifying abnormal states through dynamic thresholds includes: Point cloud segmentation: An improved PointNet++ architecture is adopted, and deformable convolutional layers are introduced into the encoder to adapt to the non-rigid and local deformation information in the cabin structure. At the same time, a multi-scale feature extraction path is constructed to capture spatial details at different scales. Foreign object classification: Based on the support vector data description method, a hypersphere model of the normal state point cloud is constructed. Normal data is tightly surrounded in the feature space, and abnormal states are judged by dynamic threshold. Points outside the threshold are marked as potential foreign objects. Confidence fusion: Stability analysis of detection results is performed in the time series dimension. Kalman filtering is used to smooth the output probability distribution of N consecutive frames, and the confidence of the presence of foreign objects in each frame is calculated. An alarm mechanism is automatically triggered when the cumulative abnormal probability over time exceeds a set value. The improved PointNet++ architecture includes: Deformable convolutional layers are introduced into the encoder to adapt to the non-rigid and local deformation information in the cabin structure. At the same time, a multi-scale feature extraction path is constructed, and 1×1×1, 3×3×3, and 5×5×5 convolutional kernels are used in the parallel branch to capture spatial details at different scales. The decoder stage introduces a dual attention mechanism of spatial and channel attention, dynamically adjusting the response intensity of each dimension in the feature map to weight and enhance the loss function. The optimized loss function is expressed as follows: Where S is the predicted point cloud output by the model. For real point clouds; This represents any point in the predicted point cloud S; Representing a true point cloud any point in it; This represents the squared Euclidean distance between point x and point y, used to quantify the positional error between the two points. This represents the squared distance from the predicted point x to the nearest point y in the true point cloud; This represents the squared distance from the true point y to the nearest point x in the predicted point cloud.

2. The method for identifying foreign objects in a battery compartment based on laser scanning and AI recognition according to claim 1, characterized in that, In the process of using a laser scanning module to scan and collect three-dimensional point cloud data inside the battery compartment at a preset cycle, the scanning path of the laser scanning module is planned based on an improved Hilbert curve model. In dense areas of the battery rack, the path density is dynamically adjusted according to the local complexity of the point cloud reconstruction model. During the scanning process, the resolution is adaptively adjusted according to the accuracy requirements of the detection task.

3. The method for identifying foreign objects in a battery compartment based on laser scanning and AI recognition according to claim 1, characterized in that, The step of preprocessing the collected raw point cloud data and converting it into a unified cabin coordinate system includes: Dynamic background modeling: Based on the adaptive Gaussian mixture model, the fixed structure and static equipment in the battery compartment are modeled. The point cloud information that has existed for a long time and has a stable shape is extracted as the baseline background. The probability distribution of each point is continuously updated to eliminate false differences caused by changes in lighting or slight vibrations. Outlier filtering: A statistical filtering algorithm is used to identify and remove spatially isolated points. The average distance from each point to its surrounding points in its local neighborhood is calculated and compared with the overall distribution. Points whose average distance from their neighborhood exceeds a preset threshold are filtered out. Data augmentation: Introducing small perturbations into the original point cloud to improve the model's adaptability to changes in pose and position; UWB positioning data processing: A density-based outlier detection method is introduced, and a multi-scale judgment strategy is constructed by integrating a statistical filtering mechanism to handle the interpolation error caused by sudden offset points in the UWB trajectory; Coordinate system transformation: Based on the pre-calibrated cabin reference points, a global coordinate mapping relationship is established. Spatial alignment of point cloud data is achieved through rotation and translation transformation matrices, and point cloud data collected at different times and angles are fused under the same spatial reference.

4. The method for identifying foreign objects in a battery compartment based on laser scanning and AI recognition according to claim 1, characterized in that, The process of acquiring personnel positioning data from the access control system, combining it with multi-time period scanning data for foreign object location tracing, and improving the alignment accuracy and trajectory estimation stability of multi-source data through dynamic time warping and spatiotemporal filtering includes: Multi-source data acquisition: Acquire personnel location data from the access control system, as well as scanning data from multiple time periods; Dynamic time warping: A dynamic time warping method based on an event triggering mechanism is adopted. Scanning nodes are triggered when personnel enter or exit a specific area, so that the trajectory time series and the laser scanning time series are non-linearly aligned on the event point set. The trajectory node alignment process is constrained by minimizing the time difference cost function. Probability density modeling: Based on the spatial interpolation algorithm unit, Kriging interpolation is introduced to perform probability density modeling on discrete point clouds, and an estimation model is constructed for the spatial location of the point cloud; Joint Kalman Filtering: For the fusion of sensors with different accuracies, the observation model combines laser point cloud and UWB positioning information, sets point cloud error and UWB error, and performs weighted processing by assigning different observation error covariances. The process noise adopts an adaptive covariance matrix to reflect the system uncertainty in positioning and motion prediction, and dynamically adjusts the filtering response capability.

5. The method for identifying foreign objects in a battery compartment based on laser scanning and AI recognition according to claim 4, characterized in that, In the multi-source data acquisition, a timestamp mapping function is established using a timestamp calibration method, and the historical data used includes personnel trajectory data and laser scan data from at least the past week.

6. A foreign object identification system for a battery compartment based on laser scanning and AI recognition, characterized in that, The system includes: The laser scanning module consists of a linear array lidar mounted on a programmable rotating bracket, used to collect three-dimensional point cloud data of the battery compartment interior at preset intervals. The data processing module, connected to the laser scanning module, is used to reduce noise in the raw point cloud data and convert it into a unified cabin coordinate system; The AI ​​recognition module, connected to the data processing module, includes a trained foreign object detection neural network, which is used to build a benchmark model based on historical normal state point cloud datasets and identify foreign objects by comparing real-time point cloud differences. The location tracking module is used to acquire personnel location data from the access control system and combine it with multi-time period scanning data to trace the location of foreign objects. The work order interface module, connected to the AI ​​recognition module and the location tracking module, is used to generate maintenance work orders that include the coordinates of foreign objects and trajectory traceability information.

7. The foreign object identification system for the battery compartment based on laser scanning and AI recognition according to claim 6, characterized in that, The data processing module includes: a point cloud preprocessing unit, used for dynamic background modeling, outlier removal from laser scanning data, and processing of abnormal interference in UWB positioning data; The coordinate system transformation unit is used to convert the raw point cloud data into a unified cabin coordinate system.

8. The foreign object identification system for the battery compartment based on laser scanning and AI recognition according to claim 6, characterized in that, The AI ​​recognition module includes: The point cloud segmentation unit adopts an improved PointNet++ architecture, introduces deformable convolutional layers in the encoder to adapt to the non-rigid and local deformation information in the cabin structure, and constructs a multi-scale feature extraction path to capture spatial details at different scales. The foreign object classification unit constructs a hypersphere model of the normal state point cloud based on the support vector data description method. It tightly surrounds the normal data in the feature space and uses a dynamic threshold to distinguish abnormal states. Points outside the threshold are marked as potential foreign objects. The confidence fusion unit is used to perform stability analysis of detection results in the time series dimension. It uses Kalman filtering to smooth the output probability distribution of N consecutive frames, calculates the confidence of the presence of foreign objects in each frame, and automatically triggers an alarm mechanism when the cumulative abnormal probability over time exceeds a set value.

Images