Intelligent robot battery safety early warning and thermal runaway suppression method and device

Abstract

The embodiment of the application provides a kind of intelligent robot battery safety early warning and thermal runaway suppression method and device, and the accurate acquisition of feature is realized by sensor array and data synchronization.Construct early warning mechanism, combine state representation and anomaly detection, establish reliable risk identification strategy.Introduce suppression optimization, through level determination and active suppression, ensure the continuous improvement of safety.The method effectively solves the shortcomings of traditional technology in feature extraction, risk identification and thermal runaway suppression, and provides technical support for robot battery safety management.

Description

Technical Field

[0001] This application relates to the field of embodied robots, specifically to a method and apparatus for battery safety early warning and thermal runaway suppression in intelligent robots. Background Technology

[0002] Existing battery safety warning methods have significant shortcomings. Traditional systems perform poorly in multi-sensor data acquisition and synchronization, failing to effectively extract accurate safety features and thus affecting the effectiveness of warnings.

[0003] Furthermore, existing technologies suffer from bottlenecks in state representation and anomaly detection. Most systems lack robust spatiotemporal attention mechanisms and fusion strategies, resulting in suboptimal risk identification accuracy.

[0004] Existing systems have technical shortcomings in thermal runaway suppression. A lack of in-depth analysis of risk levels hinders the achievement of efficient safety protection through active suppression, thus affecting the suppression effect. Solving these problems is crucial for improving the safety management capabilities of robotic batteries. Summary of the Invention

[0005] To address the problems in the existing technology, this application provides a method and apparatus for intelligent robot battery safety early warning and thermal runaway suppression, which can effectively solve the shortcomings of traditional technologies in feature extraction, risk identification and thermal runaway suppression, and provide technical support for robot battery safety management.

[0006] To solve at least one of the above problems, this application provides the following technical solution: In a first aspect, this application provides a method for intelligent robot battery safety early warning and thermal runaway suppression, including: The robot battery management system acquires temperature data from each cell collected by the temperature sensor array, voltage data from each cell collected by the voltage sensor array, and characteristic gas concentration data from the gas sensor array. Based on a unified time reference, the temperature data, the cell voltage data, and the characteristic gas concentration data are timestamped to obtain a synchronous sensing dataset. The temperature rise rate, voltage change rate, and gas concentration change rate are calculated from the synchronous sensing dataset to obtain a multimodal safety feature vector. The multimodal security feature vectors are organized into feature sequences according to time windows and input into a fusion network based on a spatiotemporal attention mechanism to perform weighted aggregation to obtain a security status representation vector. The security status representation vector is then input into a hybrid anomaly detection model to calculate anomaly scores. The anomaly scores are compared with a preset risk level threshold set to obtain a risk level determination result. Based on the risk level determination result, a corresponding warning signal is triggered and a warning message is generated and sent to the robot control system. Based on the risk level determination result, a thermal runaway suppression control instruction set is matched from the preset suppression strategy library and sent to the thermal management actuator to perform active suppression actions.

[0007] Furthermore, it also includes: obtaining raw temperature measurement data by reading the temperature values ​​at the positive and negative tab positions of each cell and the temperature value of the central region of each cell output by the temperature sensor array through the data acquisition interface; obtaining raw voltage data by reading the voltage values ​​at the terminals of each cell output by the voltage sensor array; and obtaining raw gas data by reading the electrolyte vapor concentration, carbon monoxide concentration, and hydrogen concentration values ​​output by the gas sensor array. Anomaly detection is performed on the original temperature data, the original voltage data, and the original gas data, and sampling points that exceed the preset physical reasonable range are marked as invalid. Interpolation is performed on the marked invalid sampling points, and digital filtering is performed on the completed data to remove sensor noise, thereby obtaining the temperature data, voltage data, and characteristic gas concentration data of each individual unit.

[0008] Furthermore, it also includes: generating a unified time base according to the system clock, mapping the original timestamps of the temperature measurement data of each unit, the voltage data of each unit, and the characteristic gas concentration data of each unit to the unified time base, and performing resampling interpolation on sensor data with different sampling rates so that various types of data have corresponding sampling values ​​at the same time to obtain a synchronous sensor dataset. The temperature values ​​at adjacent time points are extracted from the synchronous sensing dataset, the difference is calculated, and the difference is divided by the time interval to obtain the temperature rise rate. The voltage values ​​at adjacent time points are extracted, the difference is calculated, and the difference is divided by the time interval to obtain the voltage change rate. The gas concentration values ​​at adjacent time points are extracted, the difference is calculated, and the difference is divided by the time interval to obtain the gas concentration change rate. The temperature rise rate, the voltage change rate, and the gas concentration change rate are assembled into a multimodal safety feature vector.

[0009] Furthermore, it also includes: extracting multimodal security feature vectors from consecutive time points according to a preset window length and organizing them into a feature sequence; inputting the feature sequence into the spatial attention module of the fusion network to calculate the attention weights of each sensor channel and performing a weighted summation of the features of each channel to obtain spatial aggregated features; The spatial aggregation features are input into the temporal attention module of the fusion network to calculate the attention weights at each time step, and the features at each time step are weighted and summed to obtain the security state representation vector.

[0010] Furthermore, it also includes: inputting the security state representation vector into a deep autoencoder to perform encoding and decoding reconstruction and calculating the mean square error between the original input and the reconstruction output to obtain the reconstruction error; inputting the latent space representation of the deep autoencoder into a single-class support vector machine to calculate the function value of the distance to the decision boundary to obtain the decision function value; and weighting the reconstruction error and the decision function value according to a preset fusion weight to obtain an anomaly score. The abnormal score is compared sequentially with the attention level threshold, warning level threshold and danger level threshold in the preset risk level threshold set. The corresponding risk level is determined according to the threshold range in which the abnormal score is located, and the risk level judgment result is obtained.

[0011] Furthermore, it also includes: querying a preset warning level mapping table based on the risk level determination result to determine the warning signal type of the corresponding level; extracting the abnormality type and the location of the battery cell where the abnormality occurred from the abnormality detection process to obtain abnormality location information; and encapsulating the warning signal type and the abnormality location information according to a preset message format to obtain warning information. The warning information is sent to the robot control system through the data communication interface. The sending status of the warning information is confirmed and tracked, and warning information that has not been confirmed to be received is resent.

[0012] Furthermore, it also includes: retrieving the corresponding suppression strategy entry in the preset suppression strategy library based on the risk level determination result, and assembling the enhanced heat dissipation control parameters, power limiting parameters and electrical isolation instructions from the suppression strategy entry to obtain a thermal runaway suppression control instruction set; The enhanced heat dissipation control parameters in the thermal runaway suppression control command set are sent to the cooling fan and liquid cooling system to adjust the heat dissipation intensity. The power limiting parameters are sent to the robot drive controller to constrain the output power. The electrical isolation command is sent to the battery pack isolation switch to electrically disconnect the faulty battery from the main circuit.

[0013] Secondly, this application provides an intelligent robot battery safety early warning and thermal runaway suppression device, comprising: The battery monitoring module is used to obtain temperature measurement data of each cell collected by the temperature sensor array, voltage data of each cell collected by the voltage sensor array, and characteristic gas concentration data collected by the gas sensor array from the robot battery management system. Based on a unified time reference, the temperature measurement data, the cell voltage data, and the characteristic gas concentration data are timestamped to obtain a synchronous sensing dataset. The temperature rise rate, voltage change rate, and gas concentration change rate are calculated from the synchronous sensing dataset to obtain a multimodal safety feature vector. The safety early warning module is used to organize the multimodal safety feature vector into a feature sequence according to the time window and input it into a fusion network based on the spatiotemporal attention mechanism to perform weighted aggregation to obtain a safety state representation vector. The safety state representation vector is then input into a hybrid anomaly detection model to calculate an anomaly score. Finally, the anomaly score is compared with a preset risk level threshold set to obtain a risk level determination result. The anomaly handling module is used to trigger a warning signal of the corresponding level according to the risk level determination result and generate warning information to send to the robot control system. Based on the risk level determination result, it matches the thermal runaway suppression control instruction set from the preset suppression strategy library and sends it to the thermal management actuator to perform active suppression actions.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the intelligent robot battery safety warning and thermal runaway suppression method.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the intelligent robot battery safety warning and thermal runaway suppression method.

[0016] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the intelligent robot battery safety warning and thermal runaway suppression method.

[0017] As described above, this application provides a method and apparatus for intelligent robot battery safety early warning and thermal runaway suppression. Through sensor arrays and data synchronization, it achieves accurate feature acquisition. An early warning mechanism is constructed, combining state characterization and anomaly detection to establish a reliable risk identification strategy. Suppression optimization is introduced, ensuring continuous improvement in safety through level determination and active suppression. This method effectively addresses the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, providing technical assurance for robot battery safety management. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1This is a flowchart illustrating the intelligent robot battery safety early warning and thermal runaway suppression method in the embodiments of this application; Figure 2 This is a structural diagram of the intelligent robot battery safety early warning and thermal runaway suppression device in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0022] In view of the problems existing in the prior art, this application provides a method and apparatus for intelligent robot battery safety early warning and thermal runaway suppression. Through sensor arrays and data synchronization, it achieves accurate feature acquisition. An early warning mechanism is constructed, combining state characterization and anomaly detection to establish a reliable risk identification strategy. Suppression optimization is introduced, ensuring continuous improvement in safety through level determination and active suppression. This method effectively solves the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, providing technical assurance for robot battery safety management.

[0023] To effectively address the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, and to provide technical support for robot battery safety management, this application provides an embodiment of an intelligent robot battery safety early warning and thermal runaway suppression method. See [link to embodiment]. Figure 1 The intelligent robot battery safety early warning and thermal runaway suppression method specifically includes the following: Step S101: Obtain temperature measurement data of each cell collected by the temperature sensor array, voltage data of each cell collected by the voltage sensor array, and characteristic gas concentration data collected by the gas sensor array from the robot battery management system. Perform timestamp alignment on the temperature measurement data, the cell voltage data, and the characteristic gas concentration data based on a unified time reference to obtain a synchronous sensing dataset. Calculate the temperature rise rate, voltage change rate, and gas concentration change rate from the synchronous sensing dataset to obtain a multimodal safety feature vector. This embodiment reads multi-source sensor data output by the robot's battery management system through a data acquisition interface. The reading range covers temperature data of each cell collected by the temperature sensor array, voltage data of each cell collected by the voltage sensor array, and characteristic gas concentration data collected by the gas sensor array. The temperature sensor array has temperature measurement nodes arranged at the positive and negative tabs and the center region of each cell. The temperature values ​​at the positive and negative tabs are sensitive to localized overheating caused by internal short circuits, while the temperature value in the center region reflects the overall thermal state of the battery. The voltage sensor array samples the terminal voltage of each cell in real time, with the sampling accuracy configured to meet the detection requirements of minute voltage anomalies. The gas sensor array is deployed inside the battery pack casing and in the exhaust channel. The target gases detected include electrolyte vapor, carbon monoxide, and hydrogen. These three gases are characteristic products released from electrolyte decomposition and electrode reactions in the early stages of thermal runaway.

[0024] After acquiring the temperature data, voltage data, and characteristic gas concentration data for each individual unit, this embodiment performs outlier detection and preprocessing on the original sampled values. The outlier detection module marks sampling points that exceed a preset physical reasonable range as invalid, and performs linear interpolation to complete the invalid sampling points using adjacent valid sampling values. The completed data is then digitally filtered to remove sensor noise and electromagnetic interference. The filter uses low-pass characteristics to retain safety-related slowly varying signal components.

[0025] Accordingly, this embodiment generates a unified time base based on the system clock, and performs timestamp alignment on the preprocessed temperature measurement data, unit voltage data, and characteristic gas concentration data based on the unified time base. The timestamp alignment module maps the original timestamps of various sensor data to the unified time base. Since there are differences in the sampling rates of temperature sensors, voltage sensors, and gas sensors, the alignment process performs resampling interpolation on sensor data with different sampling rates, so that various types of data have corresponding sampled values ​​at the same time. In this embodiment, the aligned various types of sensor data are organized into a synchronous sensor dataset according to time, and each record in the synchronous sensor dataset contains the sampled values ​​of all sensor channels at the same time.

[0026] Based on the synchronous sensing dataset, this embodiment extracts sensing values ​​from adjacent time points to calculate rate of change characteristics. The temperature rise rate is calculated by extracting the temperature values ​​from adjacent time points, taking the difference, and dividing by the time interval. This temperature rise rate characteristic is valuable for early identification of thermal runaway; abnormal temperature acceleration often reaches the danger threshold before the absolute temperature. The voltage change rate is calculated by extracting the voltage values ​​from adjacent time points, taking the difference, and dividing by the time interval. Abnormal voltage drops caused by internal short circuits can be identified by negative abrupt changes in the voltage change rate. The gas concentration change rate is calculated by extracting the gas concentration values ​​from adjacent time points, taking the difference, and dividing by the time interval. A rapid increase in characteristic gas concentration indicates that a decomposition reaction may be occurring inside the battery.

[0027] Based on the aforementioned rate of change calculation results, this embodiment assembles the temperature rise rate, the voltage change rate, and the gas concentration change rate into a multimodal safety feature vector. The components of the multimodal safety feature vector are arranged sequentially according to sensor type and measurement point location, with the arrangement order corresponding to the channel numbers of the sensor array. This multimodal safety feature vector is read by subsequent step S201 and used as input data for a fusion network based on a spatiotemporal attention mechanism, to perform weighted aggregation to generate a safety state characterization vector.

[0028] Step S102: Organize the multimodal security feature vector into a feature sequence according to the time window and input it into the fusion network based on the spatiotemporal attention mechanism to perform weighted aggregation to obtain a security state representation vector. Input the security state representation vector into the hybrid anomaly detection model to calculate the anomaly score. Compare the anomaly score with the preset risk level threshold set to obtain the risk level determination result. This embodiment reads the multimodal security feature vector generated in step S101 and organizes it into a feature sequence by truncating the multimodal security feature vectors from consecutive time points according to a preset window length. The setting of the window length needs to take into account both the temporal coverage of anomaly detection and the computational response latency. The number of time points included in the window determines the depth of historical states that the fusion network can observe. This embodiment uses a sliding window method to continuously update the feature sequence. Whenever a new multimodal security feature vector arrives, the window slides forward one time point and moves the feature from the earliest time point out of the window.

[0029] After the feature sequence is organized, this embodiment inputs it into a fusion network based on a spatiotemporal attention mechanism to perform feature-level fusion. The spatial attention module of the fusion network first processes the feature sequence, calculating attention weights for the feature components of each sensor channel. The weights reflect the contribution of each sensor location to the current safety status assessment. When a local anomaly occurs at the location of a battery cell, the attention weight of the corresponding sensor channel increases accordingly. The spatial attention module performs a weighted summation of the features of each channel based on the calculated weights to obtain the spatial aggregated features.

[0030] Accordingly, in this embodiment, the spatial aggregation features are input into the temporal attention module of the fusion network for temporal correlation analysis. The temporal attention module calculates the attention weight at each time point within the window, and the weight reflects the degree of influence of the historical state at each time point on the current safety assessment. The attention weight increases at the time when a safety-related event occurs, enabling the fusion network to capture key time nodes in the abnormal evolution process. Based on the calculated weights, the temporal attention module performs a weighted summation of the spatial aggregation features at each time point to obtain a safety state representation vector. This safety state representation vector encodes the comprehensive safety state information of the battery pack within the current time window.

[0031] After the security state representation vector is generated, this embodiment inputs it into a hybrid anomaly detection model to calculate anomaly scores. The hybrid anomaly detection model comprises two components: a deep autoencoder and a single-class support vector machine. The deep autoencoder performs encoding and decoding reconstruction on the security state representation vector. The encoder compresses the input into a low-dimensional latent space representation, and the decoder reconstructs the original input from the latent space representation. This embodiment calculates the mean square error between the original input and the reconstructed output to obtain the reconstruction error. After training on normal operating data, the deep autoencoder can accurately reconstruct the representation of the normal state; however, the reconstruction error increases significantly when anomaly state representations are input.

[0032] Next, in this embodiment, the latent space representation of the deep autoencoder is input into a single-class support vector machine to calculate the decision function value. The single-class support vector machine is trained on the latent space representation of normal samples, learning the decision boundary surrounding the distribution of normal samples. The decision function value reflects the positional relationship between the current sample and the decision boundary. In this embodiment, the reconstruction error and the decision function value are weighted and summed according to a preset fusion weight to obtain an anomaly score. The anomaly score combines the reconstruction capability of the deep autoencoder and the boundary discrimination capability of the single-class support vector machine; the higher the value, the greater the degree to which the current state deviates from the normal pattern.

[0033] Based on the aforementioned anomaly score calculation results, this embodiment compares the anomaly score sequentially with each level threshold in a preset risk level threshold set. The preset risk level threshold set includes three increasing threshold boundaries: a concern level threshold, a warning level threshold, and a danger level threshold. This embodiment determines the corresponding risk level based on the threshold range in which the anomaly score falls. An anomaly score below the concern level threshold is determined to be at a normal level; between the concern level threshold and the warning level threshold; between the warning level threshold and the danger level threshold; and above the danger level threshold. The risk level determination result is read by subsequent step S201 to trigger the corresponding level's warning signal and match the thermal runaway suppression control command set.

[0034] Step S103: Trigger the corresponding level warning signal according to the risk level determination result and generate warning information to send to the robot control system. Based on the risk level determination result, match the thermal runaway suppression control instruction set from the preset suppression strategy library and send it to the thermal management actuator to perform active suppression action.

[0035] This embodiment reads the risk level determination result generated in step S102 and queries a preset warning level mapping table to determine the corresponding warning signal type. The preset warning level mapping table establishes a correspondence between risk levels and warning signal types: a focus level corresponds to enhanced status monitoring signals, a warning level corresponds to operator alarm signals, and a danger level corresponds to emergency response signals. This embodiment triggers the corresponding warning signal based on the query result. Different levels of warning signals activate different intensities of response measures and notification scopes in subsequent processing.

[0036] After the warning signal is triggered, this embodiment extracts anomaly location information from the anomaly detection process for generating the warning information. The anomaly location information includes the anomaly type and the location of the battery cell where the anomaly occurred. The anomaly type is determined by analyzing the anomaly contribution of each component in the multimodal safety feature vector, distinguishing between three types: over-temperature anomaly, voltage anomaly, and gas anomaly. The location of the battery cell where the anomaly occurred is determined by the attention weight distribution output by the spatial attention module; the battery cell corresponding to the sensor channel with the higher attention weight is the target location for anomaly localization.

[0037] Accordingly, in this embodiment, the warning signal type and the anomaly location information are encapsulated according to a preset message format to obtain warning information. The message structure of the warning information includes five fields: warning timestamp, risk level identifier, anomaly type code, anomaly unit location number, and anomaly score value. Each field is encoded with a fixed byte length for location parsing by the receiving end. In this embodiment, the encapsulated warning information is sent to the robot control system through a data communication interface. After receiving the warning information, the robot control system executes the corresponding automatic response strategy according to the risk level.

[0038] After the warning information is sent, this embodiment tracks and confirms the sending status of the warning information. The confirmation tracking module waits for a reception confirmation signal from the robot control system. If no reception confirmation signal is received within a preset timeout period, the transmission is deemed a failure, and the unconfirmed warning information is retransmitted. For warning information of a high danger level, the retransmission process simultaneously triggers a warning escalation mechanism, synchronously sending the warning information to the remote monitoring center to ensure that critical warnings are not missed.

[0039] Based on the aforementioned risk level determination results, this embodiment retrieves suppression strategy entries corresponding to the risk level from a preset suppression strategy library. The preset suppression strategy library organizes multiple suppression strategy entries according to risk level. Each suppression strategy entry includes three types of control elements: enhanced heat dissipation control parameters, power limiting parameters, and electrical isolation commands. Suppression strategy entries corresponding to the concern level only include status monitoring frequency increase parameters; suppression strategy entries corresponding to the warning level include cooling fan speed increase parameters and drive power limiting ratios; and suppression strategy entries corresponding to the danger level include emergency cooling start commands, full power cut-off commands, and fault battery isolation commands.

[0040] After the suppression strategy entries are retrieved, this embodiment reads various control parameters and instructions from them to assemble a thermal runaway suppression control instruction set. This embodiment sends the enhanced heat dissipation control parameters from the thermal runaway suppression control instruction set to the cooling fan and liquid cooling system. The cooling fan adjusts its speed according to the received speed parameters, and the liquid cooling system adjusts the coolant circulation flow rate according to the received flow parameters. Power limiting parameters are sent to the robot drive controller, which constrains the upper limit of the drive motor's output power according to the received limiting ratio. Electrical isolation instructions are sent to the battery pack isolation switch, which disconnects the faulty battery from the main circuit according to the received instructions, blocking the faulty current path to prevent the continuous release of electrical energy from the faulty battery from exacerbating the thermal runaway reaction. The execution status of the active suppression action is fed back to this embodiment by the thermal management actuator for monitoring the suppression effect and dynamically adjusting subsequent strategies.

[0041] As described above, the intelligent robot battery safety early warning and thermal runaway suppression method provided in this application embodiment can achieve accurate feature acquisition through sensor arrays and data synchronization. An early warning mechanism is constructed, combining state characterization and anomaly detection to establish a reliable risk identification strategy. Suppression optimization is introduced, ensuring continuous improvement in safety through level determination and active suppression. This method effectively solves the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, providing technical assurance for robot battery safety management.

[0042] In one embodiment of the intelligent robot battery safety early warning and thermal runaway suppression method of this application, it may further include the following: Step S201: Read the temperature values ​​at the positive and negative tabs of each cell and the temperature value of the central area of ​​each cell from the temperature sensor array through the data acquisition interface to obtain the raw temperature measurement data; read the voltage values ​​at the terminals of each cell from the voltage sensor array to obtain the raw voltage data; and read the electrolyte vapor concentration, carbon monoxide concentration, and hydrogen concentration values ​​from the gas sensor array to obtain the raw gas data. Step S202: Perform outlier detection on the original temperature data, the original voltage data, and the original gas data, and mark sampling points that exceed the preset physical reasonable range as invalid. Perform interpolation to complete the marked invalid sampling points. Perform digital filtering on the completed data to remove sensor noise and obtain temperature data, voltage data, and characteristic gas concentration data for each individual unit.

[0043] This embodiment establishes a data communication link with the robot's battery management system through a data acquisition interface, reading temperature measurement data from each cell from the temperature sensor array. The temperature sensor array deploys multiple temperature measurement nodes in each battery cell, located at the positive tab, negative tab, and center regions. The temperature measurement nodes at the positive and negative tabs utilize thermocouple sensors, which have fast response characteristics and can promptly detect localized overheating caused by internal short circuits. The temperature measurement nodes in the center region utilize thermistor sensors, which monitor the overall thermal state of the battery. This embodiment organizes the temperature values ​​read from the positive and negative tabs and the center region of each cell into raw temperature measurement data.

[0044] After the initial temperature measurement data is read, this embodiment obtains the initial voltage data by reading the terminal voltage values ​​of each individual cell from the voltage sensor array. The voltage sensor array performs parallel acquisition of the terminal voltage of each cell within the battery pack, with an acquisition accuracy configured at the millivolt level to meet the detection requirements of minute voltage anomalies. The voltage sampling clock and the temperature sampling clock share a unified reference, ensuring that the voltage data and temperature data have a corresponding relationship in the time dimension. In this embodiment, the read terminal voltage values ​​of each individual cell are organized into the initial voltage data according to the cell number order.

[0045] Accordingly, this embodiment obtains raw gas data by reading characteristic gas concentration values ​​from a gas sensor array. The gas sensor array is deployed inside the battery pack casing and at the exhaust channel location, detecting three types of target gases: electrolyte vapor, carbon monoxide, and hydrogen. Electrolyte vapor detection is achieved using a metal oxide semiconductor sensor, which has high sensitivity to organic solvent vapors. Carbon monoxide and hydrogen detection are achieved using electrochemical sensors, which selectively respond to specific gas components. In this embodiment, the read electrolyte vapor concentration values, carbon monoxide concentration values, and hydrogen concentration values ​​are organized into raw gas data.

[0046] After acquiring the raw temperature data, raw voltage data, and raw gas data, this embodiment performs outlier detection on the three types of raw data. The outlier detection module configures corresponding physical reasonable ranges for each type of sensor data. The reasonable range for temperature data is set based on the battery operating temperature range and ambient temperature conditions; the reasonable range for voltage data is set based on the nominal voltage and cutoff voltage of the battery cell; and the reasonable range for gas concentration data is set based on the sensor range and background concentration level. In this embodiment, sampling points that exceed the preset physical reasonable range are marked as invalid, and the invalidation mark serves as the trigger condition for subsequent interpolation completion.

[0047] Based on the aforementioned outlier detection results, this embodiment performs interpolation completion processing on the sampling points marked as invalid. The interpolation completion module locates the position of the invalid sampling point in the time series and extracts the adjacent valid sampling points before and after that position as the interpolation reference. This embodiment uses a linear interpolation method to calculate the completed value of the invalid sampling point. The completed value is equal to the weighted average of the values ​​of the valid sampling points before and after it, and the weight is determined by the ratio of the time interval between the invalid sampling point and the valid sampling points before and after it.

[0048] After the interpolation completion process is completed, this embodiment performs digital filtering on the completed data to remove sensor noise. The digital filtering module uses a low-pass filter to filter the data of each channel. The cutoff frequency of the low-pass filter is configured according to the frequency characteristics of the safety-related signals. The filtering process retains the slowly changing safety state signal components and suppresses high-frequency noise and electromagnetic interference components. In this embodiment, the filtered temperature data, voltage data, and gas data are output as individual unit temperature measurement data, individual unit voltage data, and characteristic gas concentration data, respectively. The individual unit temperature measurement data, individual unit voltage data, and characteristic gas concentration data are read by the timestamp alignment module in step S101 above to generate a synchronous sensing dataset.

[0049] In one embodiment of the intelligent robot battery safety early warning and thermal runaway suppression method of this application, it may further include the following: Step S301: Generate a unified time base according to the system clock, map the original timestamps of the temperature measurement data, voltage data and characteristic gas concentration data of each unit to the unified time base, and perform resampling interpolation on sensor data with different sampling rates so that all types of data have corresponding sampling values ​​at the same time to obtain a synchronous sensor dataset. Step S302: Extract the temperature values ​​at adjacent time points from the synchronous sensing dataset, calculate the difference, and divide by the time interval to obtain the temperature rise rate; extract the voltage values ​​at adjacent time points, calculate the difference, and divide by the time interval to obtain the voltage change rate; extract the gas concentration values ​​at adjacent time points, calculate the difference, and divide by the time interval to obtain the gas concentration change rate; assemble the temperature rise rate, the voltage change rate, and the gas concentration change rate into a multimodal safety feature vector.

[0050] In this embodiment, a unified time base is generated based on the system clock. The system clock uses the hardware clock of the robot's main controller as the reference source. The unified time base divides a continuous sequence of time markers with a fixed time resolution. The time base generation module converts the count value of the system clock into a timestamp format. The timestamp format includes two components: a time identifier and a time resolution. The time identifier is used to mark the position of each sampling point on the time axis.

[0051] After the unified time base is generated, this embodiment maps the original timestamps of the individual unit temperature measurement data, individual unit voltage data, and characteristic gas concentration data output in step S202 to the unified time base. The original timestamps are generated by the local clocks of each sensor acquisition module, and there are deviations and drifts between the local clocks of different sensor acquisition modules. The timestamp mapping module corrects the original timestamps according to the pre-calibrated clock deviation parameters and aligns and matches the corrected timestamps with the time stamp sequence of the unified time base.

[0052] Accordingly, this embodiment performs resampling interpolation processing on sensor data with different sampling rates. Temperature sensors, voltage sensors, and gas sensors employ different sampling rate configurations due to differences in response characteristics and monitoring requirements. Voltage data uses a higher sampling rate to capture rapid voltage fluctuations, while temperature and gas data use relatively lower sampling rates. The resampling interpolation module uses a time stamp sequence with a unified time base as the target time set and performs interpolation calculations on various types of sensor data to obtain the sampled value at the target time. This embodiment uses a linear interpolation method for resampling. For cases where the target time falls between two original sampling points, the interpolation result is obtained by weighting the time distance between the preceding and following original sampling points.

[0053] After the resampling interpolation process is completed, this embodiment organizes the various sensor data into a synchronous sensing dataset according to time. Each record in the synchronous sensing dataset corresponds to a time marker with a unified time base, and the record content includes the sampled values ​​of all sensor channels at that time. The sensor channels are arranged in the order of temperature measurement channel, voltage channel, and gas channel. The data structure of the synchronous sensing dataset ensures that subsequent feature calculations can be performed across modes based on the same time.

[0054] Based on the aforementioned synchronous sensing dataset, this embodiment extracts temperature values ​​from adjacent time points to calculate the temperature rise rate. The temperature rise rate calculation module reads the temperature channel values ​​of two consecutive records in the synchronous sensing dataset, subtracts the temperature value of the previous time point from the temperature value of the later time point to obtain the temperature difference, and divides the temperature difference by the time interval between the two time points to obtain the temperature rise rate. The temperature rise rate reflects how quickly the battery temperature changes over time, and abnormally accelerated temperature rise is a key indicator of impending thermal runaway.

[0055] Next, this embodiment extracts voltage values ​​from adjacent time points from the synchronous sensing dataset to calculate the voltage change rate. The voltage change rate calculation module uses the same method as the temperature rise rate, dividing the difference by the time interval. The voltage difference is obtained by subtracting the voltage value of the previous time point from the voltage value of the next time point, and then divided by the time interval to obtain the voltage change rate. A negative abrupt change in the voltage change rate may indicate an abnormal voltage drop caused by an internal short circuit. This embodiment also extracts gas concentration values ​​from adjacent time points from the synchronous sensing dataset to calculate the gas concentration change rate. The gas concentration change rate reflects the rate of release of characteristic gases; a rapid increase in gas concentration indicates that an electrolyte decomposition reaction may be occurring inside the battery.

[0056] After the calculations of the temperature rise rate, voltage change rate, and gas concentration change rate are completed, this embodiment assembles the three types of change rate features into a multimodal safety feature vector. The components of the multimodal safety feature vector are arranged in the following order: the temperature rise rate of each temperature measurement channel, the voltage change rate of each voltage channel, and the gas concentration change rate of each gas channel. The component arrangement order is consistent with the channel arrangement order of the synchronous sensing dataset. The multimodal safety feature vector is read by the fusion network in step S102 above and used as input data for weighted aggregation based on a spatiotemporal attention mechanism.

[0057] In one embodiment of the intelligent robot battery safety early warning and thermal runaway suppression method of this application, it may further include the following: Step S401: Extract multimodal security feature vectors from consecutive time points according to the preset window length and organize them into a feature sequence. Input the feature sequence into the spatial attention module of the fusion network to calculate the attention weights of each sensor channel and perform weighted summation on the features of each channel to obtain spatial aggregated features. Step S402: Input the spatial aggregation features into the temporal attention module of the fusion network to calculate the attention weights at each time step and perform a weighted summation on the features at each time step to obtain the security state representation vector.

[0058] This embodiment reads the multimodal security feature vector generated in step S302 and organizes it into a feature sequence by truncating the multimodal security feature vectors from consecutive time points according to a preset window length. The preset window length determines the number of historical time points covered by the feature sequence, and the setting of the window length needs to balance the temporal perception range of anomaly detection and the system response latency. This embodiment uses a sliding window mechanism to continuously update the feature sequence. When a new multimodal security feature vector is generated, the window slides forward one time point, the feature vector of the latest time point enters the tail of the window, and the feature vector of the earliest time point moves out from the head of the window.

[0059] After the feature sequence is organized, this embodiment inputs it into the spatial attention module of the fusion network for cross-channel feature fusion. The spatial attention module calculates attention weights for the feature components of each sensor channel in the feature sequence. The weight calculation process first performs a linear transformation on the features of each channel to generate query representations and key representations, and then calculates the similarity score between the query representations and key representations. In this embodiment, the similarity score is scaled to stabilize the numerical range. The scaled score is then normalized to convert it into attention weights. Normalization ensures that the sum of the attention weights of each channel is one.

[0060] Accordingly, the spatial attention module performs a weighted summation of the features of each channel based on the calculated attention weights to obtain the spatial aggregated features. The attention weights reflect the contribution of each sensor channel to the current safety status assessment. When a local anomaly occurs at the location of a battery cell, the feature components of the corresponding sensor channel deviate from the normal pattern. The spatial attention module increases the attention weight of that channel through a learned weight allocation mechanism. The spatial aggregated features generate an aggregated vector at each time step, which integrates the weighted information from all sensor channels at that time.

[0061] After the spatial aggregation features are generated, this embodiment inputs them into the temporal attention module of the fusion network for cross-time-matter feature fusion. The temporal attention module calculates attention weights for the spatial aggregation features at each time step within the window, using a query key matching mechanism similar to that of the spatial attention module. The temporal attention module performs a linear transformation on the spatial aggregation features at each time step to generate temporal query representations and temporal key representations, calculates the similarity score between the temporal query representations and temporal key representations, and performs scaling and normalization processing to obtain the attention weights at each time step.

[0062] Based on the aforementioned time attention weight calculation results, the time attention module performs a weighted summation of the spatial aggregation features at each time step to obtain a safety state representation vector. The time attention weight reflects the degree of influence of historical time-step states on the current safety assessment. The spatial aggregation features at the time of a safety-related event contain anomalous signal components. The time attention module strengthens the anomalous signal during the aggregation process by increasing the attention weight at that time. The safety state representation vector encodes the comprehensive safety state information of the battery pack within the current time window, fusing multi-sensor information in the spatial dimension with historical evolution information in the temporal dimension. This safety state representation vector is read by the hybrid anomaly detection model in step S102 and serves as input data for the deep autoencoder and single-class support vector machine to perform anomaly scoring calculations.

[0063] In one embodiment of the intelligent robot battery safety early warning and thermal runaway suppression method of this application, it may further include the following: Step S501: Input the security state representation vector into a deep autoencoder to perform encoding and decoding reconstruction and calculate the mean square error between the original input and the reconstruction output to obtain the reconstruction error. Input the latent space representation of the deep autoencoder into a single-class support vector machine to calculate the function value of the distance to the decision boundary to obtain the decision function value. Sum the reconstruction error and the decision function value according to the preset fusion weight to obtain the anomaly score. Step S502: The abnormal score is compared sequentially with the attention level threshold, warning level threshold and danger level threshold in the preset risk level threshold set, and the corresponding risk level is determined according to the threshold range in which the abnormal score is located to obtain the risk level judgment result.

[0064] This embodiment reads the security state representation vector generated in step S402 and inputs it into a deep autoencoder for encoding and decoding reconstruction. The deep autoencoder adopts a symmetrical network structure of encoder and decoder. The encoder part consists of a multi-layer fully connected network, and the dimension of each hidden layer decreases layer by layer along the encoding direction to achieve feature compression. The encoder receives the security state representation vector as input, and outputs a low-dimensional latent space representation after layer-by-layer nonlinear transformation. The latent space representation retains the key information components in the security state representation vector.

[0065] After the latent space representation is generated, the decoder part of the deep autoencoder performs reconstruction processing on it. The network structure of the decoder part is symmetrical to that of the encoder, and the dimension of each hidden layer increases layer by layer along the decoding direction to achieve feature reconstruction. The decoder receives the latent space representation as input, performs layer-by-layer nonlinear transformation, and outputs a reconstructed vector. The dimension of the reconstructed vector is the same as that of the original safe state representation vector. During the offline training phase, the deep autoencoder uses the safe state representation vector under normal operating conditions as training data, and the training objective is to minimize the difference between the original input and the reconstructed output.

[0066] Accordingly, this embodiment calculates the mean square error between the original safe state representation vector and the reconstructed vector to obtain the reconstruction error. The mean square error is calculated by squaring the differences between corresponding components of the original vector and the reconstructed vector and then averaging the results. The reconstruction error reflects the accuracy of the deep autoencoder's reconstruction of the current input. After training on normal data, the deep autoencoder can accurately reconstruct the representation vector of the normal state. However, when the input is a safe state representation vector of an abnormal state, the abnormal pattern deviates from the distribution of the training data, leading to a decrease in reconstruction accuracy and a corresponding increase in reconstruction error.

[0067] After the reconstruction error calculation is completed, this embodiment inputs the latent space representation of the deep autoencoder into a single-class support vector machine to calculate the decision function value. During the offline training phase, the single-class support vector machine uses the latent space representation of normal samples as training data to learn the decision boundary surrounding the distribution of normal samples. The decision boundary forms a hypersphere structure in the latent space, with normal samples distributed inside the hypersphere and abnormal samples distributed outside. This embodiment calculates the decision function value by measuring the distance of the current latent space representation from the decision boundary. A positive decision function value indicates that the sample is located outside the boundary, and a negative decision function value indicates that the sample is located inside the boundary. The absolute value reflects the distance of the sample from the boundary.

[0068] Based on the aforementioned reconstruction error and decision function value, this embodiment calculates an anomaly score by weighting and summing the two according to a preset fusion weight. The preset fusion weight is determined based on the anomaly detection performance of the deep autoencoder and the single-class support vector machine on the validation dataset. The configuration of the fusion weight allows the advantages of the two detection methods to complement each other. The anomaly score combines the anomaly measurement based on reconstruction error of the deep autoencoder and the anomaly measurement based on boundary distance of the single-class support vector machine. A higher anomaly score indicates a greater deviation of the current safety state from the normal pattern.

[0069] After the anomaly score is calculated, this embodiment compares it sequentially with the thresholds of each level in a preset risk level threshold set. The preset risk level threshold set includes three threshold boundaries arranged in ascending order: a concern level threshold, a warning level threshold, and a danger level threshold. These three thresholds divide the numerical range of the anomaly score into four intervals.

[0070] This embodiment first compares the anomaly score with the attention level threshold. If the anomaly score is lower than the attention level threshold, it is determined to be at the normal level. If the anomaly score is not lower than the attention level threshold, it is further compared with the warning level threshold. If the anomaly score is between the attention level threshold and the warning level threshold, it is determined to be at the attention level. If the anomaly score is not lower than the warning level threshold, it is further compared with the danger level threshold. If the anomaly score is between the warning level threshold and the danger level threshold, it is determined to be at the warning level; if the anomaly score is not lower than the danger level threshold, it is determined to be at the danger level. The risk level determination result is read by the aforementioned step S103 and used to trigger the corresponding level of warning signal and the matching thermal runaway suppression control instruction set.

[0071] In one embodiment of the intelligent robot battery safety early warning and thermal runaway suppression method of this application, it may further include the following: Step S601: Based on the risk level determination result, query the preset warning level mapping table to determine the warning signal type of the corresponding level, extract the abnormality type and the location of the battery cell where the abnormality occurred from the abnormality detection process to obtain the abnormality location information, and encapsulate the warning signal type and the abnormality location information according to the preset message format to obtain the warning information; Step S602: Send the warning information to the robot control system through the data communication interface, confirm and track the sending status of the warning information, and resend the warning information if it is not confirmed to be received.

[0072] This embodiment reads the risk level determination result generated in step S502 above, and queries a preset warning level mapping table to determine the warning signal type corresponding to the risk level. The preset warning level mapping table uses the risk level as the index key and stores the warning signal type and response intensity configuration corresponding to each risk level. The normal level corresponds to an empty warning signal in the mapping table, triggering no warning output. The attention level corresponds to an enhanced status monitoring signal, which triggers an increase in monitoring frequency and continuous tracking of abnormal development trends. The warning level corresponds to an operator alarm signal, which triggers audible and visual alarm prompts and power limiting measures. The danger level corresponds to an emergency response signal, which triggers emergency shutdown, electrical isolation, and emergency cooling activation.

[0073] After the warning signal type is determined, this embodiment extracts anomaly location information from the anomaly detection process. Anomaly type extraction is achieved by analyzing the anomaly contribution of each feature component in the multimodal safety feature vector. This embodiment calculates the degree to which each feature component deviates from the normal range and identifies the feature category with the highest contribution. The highest contribution of the temperature rise rate component indicates an over-temperature anomaly; the highest contribution of the voltage change rate component indicates a voltage anomaly; and the highest contribution of the gas concentration change rate component indicates a gas anomaly. The location of the battery cell where the anomaly occurred is determined by the attention weight distribution output by the spatial attention module. This embodiment extracts the sensor channel number with the highest attention weight and maps it to the corresponding battery cell location number.

[0074] Accordingly, this embodiment encapsulates the warning signal type and the anomaly location information according to a preset message format to obtain warning information. The preset message format defines the field structure and byte allocation scheme for each field of the warning information. The message header is a warning timestamp field recording the time when the warning was generated, followed by a risk level identifier field, a warning signal type field, an anomaly type code field, an anomaly unit location number field, and an anomaly score value field. Each field is encoded with a fixed byte length, and no separators are used between fields. The receiving end locates and parses the content of each field based on the field offset. In this embodiment, the values ​​of each field are written into the message buffer in the message format order to complete the encapsulation of the warning information.

[0075] After the warning information is encapsulated, this embodiment sends it to the robot control system via a data communication interface. The data communication interface establishes a communication link with the control system using the robot's internal bus protocol, and the communication link supports a reliable transmission mode to ensure the delivery of the warning information. In this embodiment, the encapsulated warning information is written into a sending queue, and the data communication interface reads the warning information from the sending queue, performs protocol encapsulation, and then sends it to the robot control system.

[0076] Based on the aforementioned warning information sending action, this embodiment tracks and confirms the sending status of the warning information. After the warning information is sent, the confirmation tracking module starts a confirmation waiting timer, waiting for a reception confirmation signal from the robot control system. After successfully receiving and parsing the warning information, the robot control system generates a reception confirmation signal and sends it back to this embodiment. Upon receiving the reception confirmation signal, the confirmation tracking module marks the corresponding warning information as successfully sent and clears the confirmation waiting status.

[0077] During the confirmation tracking process, this embodiment performs retransmission processing on unacknowledged early warning information. If the confirmation waiting timer expires without receiving a confirmation signal, the confirmation tracking module determines that the transmission has failed and triggers the retransmission process. The retransmission process rewrites the unacknowledged early warning information into the transmission queue and increments the retransmission counter. The data communication interface then retransmits the retransmitted early warning information. This embodiment sets an upper limit on the number of retransmissions. When the retransmission counter reaches the upper limit and no confirmation is received, the early warning information is marked as a transmission error and an error log is recorded.

[0078] In one embodiment of the intelligent robot battery safety early warning and thermal runaway suppression method of this application, it may further include the following: Step S701: Based on the risk level determination result, retrieve the corresponding risk level suppression strategy entry in the preset suppression strategy library, and assemble the enhanced heat dissipation control parameters, power limiting parameters and electrical isolation instructions from the suppression strategy entry to obtain the thermal runaway suppression control instruction set; Step S702: Send the enhanced heat dissipation control parameters from the thermal runaway suppression control command set to the cooling fan and liquid cooling system to adjust the heat dissipation intensity; send the power limiting parameters to the robot drive controller to constrain the output power; and send the electrical isolation command to the battery pack isolation switch to electrically disconnect the faulty battery from the main circuit.

[0079] This embodiment reads the risk level determination result generated in step S502 and retrieves the corresponding suppression strategy entry from the preset suppression strategy library based on the risk level determination result. The preset suppression strategy library organizes multiple suppression strategy entries according to risk level, and each suppression strategy entry is indexed and associated with a specific risk level. This embodiment uses the risk level determination result as the search key to locate the matching suppression strategy entry in the preset suppression strategy library and read its content.

[0080] After the suppression strategy entries are retrieved, this embodiment reads the enhanced heat dissipation control parameters. The enhanced heat dissipation control parameters consist of two components: the cooling fan speed setting and the liquid cooling system flow rate setting. Different risk levels correspond to different intensities of heat dissipation control parameter configurations. For suppression strategy entries at the "attention" level, the enhanced heat dissipation control parameters maintain the normal heat dissipation configuration without adjustment. For suppression strategy entries at the "warning" level, the cooling fan speed setting is increased to medium intensity, and the liquid cooling system flow rate setting is increased accordingly. For suppression strategy entries at the "danger" level, the cooling fan speed setting is increased to maximum intensity, and the liquid cooling system flow rate setting is adjusted to the emergency cooling level.

[0081] Accordingly, this embodiment reads the power limiting parameter from the suppression strategy entry. The power limiting parameter defines the maximum percentage of power that the robot drive system is allowed to output, and the power limiting percentage is expressed as a percentage of the rated power. In the suppression strategy entry for the concern level, the power limiting parameter is not limited; in the suppression strategy entry for the warning level, the power limiting parameter constrains the output power of the drive system to not exceed a preset percentage of the rated power; in the suppression strategy entry for the danger level, the power limiting parameter is set to zero to achieve complete power cutoff.

[0082] After the power limiting parameters are read, this embodiment reads the electrical isolation command from the suppression strategy entries. The electrical isolation command defines whether to perform an electrical disconnection action between the faulty battery and the main circuit; the command content includes an isolation action enable flag and the target isolating switch number. In the suppression strategy entries for concern and warning levels, the isolation action enable flag of the electrical isolation command is disabled; in the suppression strategy entries for danger levels, the isolation action enable flag of the electrical isolation command is enabled. This embodiment assembles the read enhanced heat dissipation control parameters, power limiting parameters, and electrical isolation commands to obtain a thermal runaway suppression control command set.

[0083] Based on the aforementioned thermal runaway suppression control instruction set, this embodiment sends the enhanced heat dissipation control parameters to the cooling fan and liquid cooling system to adjust the heat dissipation intensity. The cooling fan controller receives the cooling fan speed setpoint and adjusts the fan motor drive signal, increasing or decreasing the fan speed according to the setpoint. The liquid cooling system controller receives the liquid cooling system flow rate setpoint and adjusts the operating status of the circulation pump and the opening of the flow control valve, adjusting the coolant circulation flow rate accordingly. This increased heat dissipation intensity accelerates heat dissipation from the battery pack and slows down the temperature rise.

[0084] Next, in this embodiment, the power limiting parameters are sent to the robot drive controller to perform output power constraint. After receiving the power limiting parameters, the robot drive controller modifies its internal power output upper limit configuration, and the output power of the drive motor is constrained within the range defined by the power limiting parameters. Power limiting reduces the discharge current intensity of the battery pack, reduces Joule heat generation inside the battery, and suppresses temperature rise from the heat source.

[0085] After the power limiting is executed, this embodiment sends an electrical isolation command to the battery pack isolating switch to electrically disconnect the faulty battery from the main circuit. Upon receiving the electrical isolation command, the battery pack isolating switch checks the isolation action enable flag. If the enable flag is active, it executes the disconnection action of the corresponding switch according to the target isolating switch number. The disconnection action of the isolating switch severs the electrical connection between the faulty battery and the battery pack main circuit, blocking the fault current path to prevent the continuous release of electrical energy from the faulty battery from exacerbating the thermal runaway reaction. The three types of suppression actions—enhanced heat dissipation control, power limiting, and electrical isolation—form a multi-layered thermal runaway suppression system. The execution status of each suppression action is fed back to this embodiment by the corresponding actuator for monitoring the suppression effect and dynamically adjusting subsequent strategies.

[0086] To effectively address the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, and to provide technical support for robot battery safety management, this application provides an embodiment of an intelligent robot battery safety early warning and thermal runaway suppression device for implementing all or part of the aforementioned intelligent robot battery safety early warning and thermal runaway suppression method. See [link to embodiment]. Figure 2 The intelligent robot battery safety early warning and thermal runaway suppression device specifically includes the following components: The battery monitoring module 10 is used to obtain temperature measurement data of each cell collected by the temperature sensor array, voltage data of each cell collected by the voltage sensor array, and characteristic gas concentration data collected by the gas sensor array from the robot battery management system. Based on a unified time reference, the temperature measurement data, the cell voltage data, and the characteristic gas concentration data are timestamped to obtain a synchronous sensing dataset. The temperature rise rate, voltage change rate, and gas concentration change rate are calculated from the synchronous sensing dataset to obtain a multimodal safety feature vector. The safety early warning module 20 is used to organize the multimodal safety feature vector into a feature sequence according to the time window and input it into a fusion network based on the spatiotemporal attention mechanism to perform weighted aggregation to obtain a safety state representation vector. The safety state representation vector is then input into a hybrid anomaly detection model to calculate an anomaly score. The anomaly score is then compared with a preset risk level threshold set to obtain a risk level determination result. The anomaly handling module 30 is used to trigger a warning signal of the corresponding level according to the risk level determination result and generate warning information to send to the robot control system. Based on the risk level determination result, it matches the thermal runaway suppression control instruction set from the preset suppression strategy library and sends it to the thermal management actuator to perform active suppression action.

[0087] As described above, the intelligent robot battery safety early warning and thermal runaway suppression device provided in this application embodiment can achieve accurate feature acquisition through sensor arrays and data synchronization. An early warning mechanism is constructed, combining state characterization and anomaly detection to establish a reliable risk identification strategy. Suppression optimization is introduced, ensuring continuous improvement in safety through level determination and active suppression. This method effectively solves the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, providing technical assurance for robot battery safety management.

[0088] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the intelligent robot battery safety early warning and thermal runaway suppression method.

[0089] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described intelligent robot battery safety early warning and thermal runaway suppression method.

[0090] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-mentioned intelligent robot battery safety early warning and thermal runaway suppression method.

[0091] In this embodiment of the invention, precise feature acquisition is achieved through sensor arrays and data synchronization. An early warning mechanism is constructed, combining state characterization and anomaly detection to establish a reliable risk identification strategy. Suppression optimization is introduced, ensuring continuous improvement in safety through level determination and active suppression. This method effectively addresses the shortcomings of traditional technologies in feature extraction, risk identification, and thermal runaway suppression, providing technical assurance for robot battery safety management.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent robot battery safety early warning and thermal runaway suppression, characterized in that, The method includes: The robot battery management system acquires temperature data from each cell collected by the temperature sensor array, voltage data from each cell collected by the voltage sensor array, and characteristic gas concentration data from the gas sensor array. Based on a unified time reference, the temperature data, the cell voltage data, and the characteristic gas concentration data are timestamped to obtain a synchronous sensing dataset. The temperature rise rate, voltage change rate, and gas concentration change rate are calculated from the synchronous sensing dataset to obtain a multimodal safety feature vector. The multimodal security feature vectors are organized into feature sequences according to time windows and input into a fusion network based on a spatiotemporal attention mechanism to perform weighted aggregation to obtain a security status representation vector. The security status representation vector is then input into a hybrid anomaly detection model to calculate anomaly scores. The anomaly scores are compared with a preset risk level threshold set to obtain a risk level determination result. Based on the risk level determination result, a corresponding warning signal is triggered and a warning message is generated and sent to the robot control system. Based on the risk level determination result, a thermal runaway suppression control instruction set is matched from the preset suppression strategy library and sent to the thermal management actuator to perform active suppression actions.

2. The intelligent robot battery safety early warning and thermal runaway suppression method according to claim 1, characterized in that, The acquisition of individual cell temperature data from the temperature sensor array, individual cell voltage data from the voltage sensor array, and characteristic gas concentration data from the gas sensor array from the robot battery management system includes: The raw temperature data is obtained by reading the temperature values ​​at the positive and negative tabs of each cell and the temperature value of the central area of ​​each cell from the temperature sensor array output by the data acquisition interface; the raw voltage data is obtained by reading the voltage values ​​at the terminals of each cell from the voltage sensor array output by the voltage sensor array; and the raw gas data is obtained by reading the electrolyte vapor concentration, carbon monoxide concentration, and hydrogen concentration values ​​from the gas sensor array output by the gas sensor array. Anomaly detection is performed on the original temperature data, the original voltage data, and the original gas data, and sampling points that exceed the preset physical reasonable range are marked as invalid. Interpolation is performed on the marked invalid sampling points, and digital filtering is performed on the completed data to remove sensor noise, thereby obtaining the temperature data, voltage data, and characteristic gas concentration data of each individual unit.

3. The intelligent robot battery safety early warning and thermal runaway suppression method according to claim 1, characterized in that, The process involves aligning the temperature measurement data, the unit voltage data, and the characteristic gas concentration data with timestamps based on a unified time reference to obtain a synchronized sensing dataset. From this synchronized sensing dataset, the temperature rise rate, voltage change rate, and gas concentration change rate are calculated to obtain a multimodal safety feature vector, including: A unified time base is generated based on the system clock. The original timestamps of the temperature measurement data, voltage data, and characteristic gas concentration data of each individual are mapped to the unified time base. Resampling interpolation is performed on sensor data with different sampling rates so that all types of data have corresponding sampling values ​​at the same time to obtain a synchronous sensor dataset. The temperature values ​​at adjacent time points are extracted from the synchronous sensing dataset, the difference is calculated, and the difference is divided by the time interval to obtain the temperature rise rate. The voltage values ​​at adjacent time points are extracted, the difference is calculated, and the difference is divided by the time interval to obtain the voltage change rate. The gas concentration values ​​at adjacent time points are extracted, the difference is calculated, and the difference is divided by the time interval to obtain the gas concentration change rate. The temperature rise rate, the voltage change rate, and the gas concentration change rate are assembled into a multimodal safety feature vector.

4. The intelligent robot battery safety early warning and thermal runaway suppression method according to claim 1, characterized in that, The step of organizing the multimodal security feature vectors into a feature sequence according to a temporal window and inputting it into a fusion network based on a spatiotemporal attention mechanism to perform weighted aggregation to obtain a security state representation vector includes: The multimodal security feature vectors of consecutive moments are extracted according to the preset window length and organized into a feature sequence. The feature sequence is input into the spatial attention module of the fusion network to calculate the attention weight of each sensor channel and perform weighted summation on the features of each channel to obtain the spatial aggregated features. The spatial aggregation features are input into the temporal attention module of the fusion network to calculate the attention weights at each time step, and the features at each time step are weighted and summed to obtain the security state representation vector.

5. The intelligent robot battery safety early warning and thermal runaway suppression method according to claim 1, characterized in that, The step of inputting the safety status representation vector into the hybrid anomaly detection model to calculate an anomaly score, and comparing the anomaly score with a preset risk level threshold set to obtain a risk level determination result includes: The security state representation vector is input into a deep autoencoder to perform encoding and decoding reconstruction, and the mean square error between the original input and the reconstruction output is calculated to obtain the reconstruction error. The latent space representation of the deep autoencoder is input into a single-class support vector machine to calculate the function value of the distance to the decision boundary to obtain the decision function value. The reconstruction error and the decision function value are weighted and summed according to a preset fusion weight to obtain the anomaly score. The abnormal score is compared sequentially with the attention level threshold, warning level threshold and danger level threshold in the preset risk level threshold set. The corresponding risk level is determined according to the threshold range in which the abnormal score is located, and the risk level judgment result is obtained.

6. The intelligent robot battery safety early warning and thermal runaway suppression method according to claim 1, characterized in that, The step of triggering a corresponding level of early warning signal based on the risk level determination result and generating early warning information to send to the robot control system includes: Based on the risk level determination result, the corresponding warning signal type is determined by querying the preset warning level mapping table. The abnormality type and the location of the battery cell where the abnormality occurred are extracted from the abnormality detection process to obtain the abnormality location information. The warning signal type and the abnormality location information are encapsulated according to the preset message format to obtain the warning information. The warning information is sent to the robot control system through the data communication interface. The sending status of the warning information is confirmed and tracked, and warning information that has not been confirmed to be received is resent.

7. The intelligent robot battery safety early warning and thermal runaway suppression method according to claim 1, characterized in that, The step of matching a thermal runaway suppression control instruction set from a preset suppression strategy library based on the risk level determination result and sending it to the thermal management actuator to execute active suppression actions includes: Based on the risk level determination result, the corresponding risk level suppression strategy entry is retrieved from the preset suppression strategy library. The enhanced heat dissipation control parameters, power limiting parameters and electrical isolation instructions are read from the suppression strategy entry to assemble a thermal runaway suppression control instruction set. The enhanced heat dissipation control parameters in the thermal runaway suppression control command set are sent to the cooling fan and liquid cooling system to adjust the heat dissipation intensity. The power limiting parameters are sent to the robot drive controller to constrain the output power. The electrical isolation command is sent to the battery pack isolation switch to electrically disconnect the faulty battery from the main circuit.

8. A smart robot battery safety early warning and thermal runaway suppression device, characterized in that, The device includes: The battery monitoring module is used to obtain temperature measurement data of each cell collected by the temperature sensor array, voltage data of each cell collected by the voltage sensor array, and characteristic gas concentration data collected by the gas sensor array from the robot battery management system. Based on a unified time reference, the temperature measurement data, the cell voltage data, and the characteristic gas concentration data are timestamped to obtain a synchronous sensing dataset. The temperature rise rate, voltage change rate, and gas concentration change rate are calculated from the synchronous sensing dataset to obtain a multimodal safety feature vector. The safety early warning module is used to organize the multimodal safety feature vector into a feature sequence according to the time window and input it into a fusion network based on the spatiotemporal attention mechanism to perform weighted aggregation to obtain a safety state representation vector. The safety state representation vector is then input into a hybrid anomaly detection model to calculate an anomaly score. Finally, the anomaly score is compared with a preset risk level threshold set to obtain a risk level determination result. The anomaly handling module is used to trigger a warning signal of the corresponding level according to the risk level determination result and generate warning information to send to the robot control system. Based on the risk level determination result, it matches the thermal runaway suppression control instruction set from the preset suppression strategy library and sends it to the thermal management actuator to perform active suppression actions.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent robot battery safety early warning and thermal runaway suppression method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the intelligent robot battery safety early warning and thermal runaway suppression method according to any one of claims 1 to 7.

Images