A method and device for early warning of thermal runaway in MiniLED displays based on a coupling model.

Abstract

This invention provides a method and device for early warning of thermal runaway in MiniLED displays based on a coupling model. It involves activating multi-scale sensing nodes on the MiniLED display to collect continuous sampling records of chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operating cycle. This yields a multi-scale photoelectric and thermal time-series dataset anchored to the conduction links. Based on this dataset, a causal convolutional network is used to mine the conduction causal chains of photoelectric and thermal interactions between different stages of thermal runaway. By using reverse causal verification logic to remove pseudo-correlated sampling records, a set of cross-stage coupling-dominant causal chains is obtained. This set is then subjected to bidirectional feedback calibration to obtain a feedback-corrected coupled conduction causal network. This network drives a thermal coupling evolution model, generating a coupled-driven thermal operation evolution probability map and a dynamically updated instability critical region as the derivation process progresses. Combined with real-time sampling records from the MiniLED display, conduction causal chain matching and early warning judgment are performed to provide thermal runaway early warning. This invention effectively enhances the risk prevention and control capabilities of MiniLED thermal runaway.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and device for early warning of thermal runaway in MiniLED displays based on a coupling model. Background Technology

[0002] With the rapid popularization of MiniLED display technology, thermal runaway early warning technology for MiniLED displays has become crucial for ensuring stable equipment operation and extending product lifespan. Currently, most commonly used thermal runaway early warning methods in the industry focus on monitoring specific types of heat-related state information. They judge the risk of thermal runaway by setting static risk thresholds. Some technologies construct evolutionary models based on general thermophysical formulas to extrapolate thermal states and assist in early warning judgment. However, these existing methods generally suffer from insufficient correlation between monitoring data and the photoelectric and thermal conduction processes inside the display screen. They cannot accurately match the characteristics of multi-node coupling and conduction of MiniLED chips, substrates, and packages. The judgment logic based on static thresholds is difficult to adapt to the dynamic process of the gradual development of thermal runaway. The evolutionary models constructed from general thermophysical formulas deviate from the actual operating state under the differences in actual production and manufacturing and the disturbance of the usage environment. This makes it difficult to meet the accuracy and foresight of early warning judgments, posing potential risks to the stable operation of MiniLED displays. Summary of the Invention

[0003] In view of this, the present invention provides a method and device for early warning of thermal runaway in MiniLED displays based on a coupling model. The technical solution of the present invention is implemented as follows:

[0004] On one hand, embodiments of the present invention provide a method for early warning of thermal runaway in MiniLED displays based on a coupling model. This method includes: synchronously activating multi-scale sensing nodes of the MiniLED display to collect continuous sampling records of chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operating cycle; and using the coupling conduction link marking logic built into the sensing nodes to bind a corresponding chip-substrate-package coupling node conduction link mark to each continuous sampling record, thereby obtaining a multi-scale opto-thermal time-series dataset anchored to the conduction link; based on the multi-scale opto-thermal time-series dataset anchored to the conduction link, mining the conduction causal chain of opto-thermal interaction between different stages of thermal runaway using a causal convolutional network; and stripping pseudo-correlated sampling records using reverse causal verification logic to obtain cross-stage... The system employs a set of coupled dominant causal chains as its core input. Continuous sampling records of chip manufacturing discreteness and environmental thermal disturbances are introduced for bidirectional mutual feedback calibration. Causal chain deviation propagation identification logic removes the propagation deviation causal chains generated during calibration, resulting in a mutually corrected coupled propagation causal network. This corrected network drives a thermal coupling evolution model, performing dynamic extrapolation of multi-stage photoelectric thermal conduction. This generates a coupled-driven thermal operation evolution probability map and a dynamically updated instability critical domain. Based on the coupled-driven thermal operation evolution probability map and the dynamically instability critical domain, combined with real-time sampling records from the MiniLED display, a propagation causal chain matching and early warning judgment is performed to complete the thermal runaway early warning for the MiniLED display.

[0005] On the other hand, embodiments of the present invention provide an early warning device, comprising: a data acquisition module, used to synchronously initiate multi-scale sensing nodes of a MiniLED display to collect continuous sampling records of chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operating cycle, and to bind a corresponding chip-substrate-package coupling node's conduction link label to each continuous sampling record through coupling conduction link labeling logic built into the sensing nodes, thereby obtaining a multi-scale opto-thermal time-series dataset anchored to the conduction links; a causal mining module, used to mine conduction causal chains of opto-thermal effects between different thermal runaway process stages based on the multi-scale opto-thermal time-series dataset anchored to the conduction links, using a causal convolutional network, and to remove pseudo-correlated sampling records through reverse causal verification logic, thereby obtaining a set of cross-stage coupling dominant causal chains; and a mutual feedback calibration module. The system is divided into four modules: a core input module and a dynamic simulation module. The core input module uses a set of cross-stage coupled causal chains as the main input, and continuously samples the chip manufacturing discreteness and environmental thermal disturbances for bidirectional mutual feedback calibration. The simulation module removes the conduction deviation causal chains generated during the calibration process through the causal chain deviation identification logic, resulting in a mutually fed-back corrected coupled conduction causal network. The dynamic simulation module drives the thermal coupling evolution model through the mutually fed-back corrected coupled conduction causal network to perform dynamic simulation of multi-stage photoelectric thermal conduction, generating a coupled-driven thermal operation evolution probability map and a dynamic instability critical domain that is updated in real time as the simulation progresses. The runaway early warning module is used to perform conduction causal chain matching and early warning judgment based on the coupled-driven thermal operation evolution probability map and the dynamic instability critical domain, combined with the real-time sampling records of the MiniLED display, to complete the thermal runaway early warning of the MiniLED display. Attached Figure Description

[0006] Figure 1 This is a schematic diagram illustrating the implementation process of a thermal runaway early warning method for a MiniLED display screen based on a coupling model, provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of the composition of an early warning device provided in an embodiment of the present invention. Detailed Implementation

[0008] This invention provides a method for early warning of thermal runaway in a MiniLED display screen based on a coupling model, which can be executed by the processor of the display screen.

[0009] Figure 1 This is a schematic diagram illustrating the implementation process of a thermal runaway early warning method for a MiniLED display screen based on a coupling model, as provided in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0010] Step S100: Simultaneously start the MiniLED display multi-scale sensing node to collect continuous sampling records of chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operating cycle. Then, through the coupling conduction link marking logic built into the sensing node, bind the corresponding chip-substrate-package coupling node conduction link mark to each continuous sampling record to obtain a multi-scale optoelectronic thermal time series dataset anchored to the conduction link.

[0011] Chip junction temperature fluctuation refers to the change in chip junction temperature over time throughout the entire operating cycle of a MiniLED display, reflecting the chip's heat dissipation characteristics under different operating conditions. For example, when displaying a high-brightness image, the chip's workload increases, and the junction temperature may rise; conversely, when displaying a low-brightness image, the junction temperature may decrease. Electro-optical conversion efficiency decay is the phenomenon where the efficiency of the MiniLED chip in converting electrical energy into light energy gradually decreases with increasing operating time. Substrate thermal conductivity differences refer to the variations in thermal conductivity at different locations or regions of the substrate in a MiniLED display. Due to non-uniformity in the substrate material or differences in structural design, the speed and path of heat conduction on the substrate will differ.

[0012] Multi-scale sensing nodes are devices used to collect the aforementioned data, enabling monitoring of chips, substrates, and other components at different scales. For example, at a microscopic scale, the junction temperature of a single chip can be monitored, while at a macroscopic scale, the thermal conductivity of the entire display substrate area can be monitored. The sensing node's built-in coupling conduction link marking logic is a preset rule or algorithm used to assign a corresponding chip-substrate-package coupling node conduction link mark to each continuous sampling record. The conduction link mark is identification information used to identify different conduction paths, clearly showing the specific conduction link corresponding to each sampling record.

[0013] In one implementation, step S100 may specifically include the following steps S110 to S170:

[0014] Step S110: Simultaneously start the multi-scale sensing nodes of the MiniLED display to collect continuous sampling records of chip junction temperature fluctuation, electro-optical conversion efficiency decay, and substrate thermal conduction differences according to the time sequence nodes of the entire operation cycle. Bind each continuous sampling record to the corresponding time sequence node's operation status record one-to-one to generate a time-bound sampling record sequence containing time sequence markers and operation status association information.

[0015] The operational status log contains information about the display's operating status at each time point, including the brightness, color mode, and frame rate of the displayed image. This information helps in subsequent analysis of the relationship between the sampled data and the display's operational status.

[0016] In this embodiment of the invention, the multi-scale sensing nodes collect data according to the time sequence nodes of the entire operating cycle. For example, at each minute-by-minute time sequence node, the node simultaneously collects continuous sampling records of chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conductivity differences. Then, each segment of the collected continuous sampling record is bound one-to-one with the corresponding operating status record of that time sequence node. In this way, each segment of the generated time-bound sampling record sequence contains a corresponding time sequence marker and operating status association information.

[0017] Step S120: Perform conduction link tracing on the time-bound sampling record sequence, trace the complete conduction path of the chip-substrate-package coupling node corresponding to each continuous sampling record, and generate conduction path tracing results containing detailed information on the start and end nodes of the path.

[0018] The complete conduction path of the chip-substrate-package coupling node is the specific path through which heat, current, and other substances are transferred between the chip, substrate, and package. The start and end nodes of the path represent the beginning and end positions of the conduction path, respectively. Detailed information includes the specific location and physical characteristics of each node. Tracing the source can be achieved using data analysis and modeling methods. For example, first, a heat conduction model of the chip-substrate-package coupling node is established, which includes information such as the physical characteristics and geometry of the chip, substrate, and package. Then, based on data from each continuous sampling record, such as chip junction temperature fluctuations and substrate thermal conductivity differences, the model is used to reverse-engineer the flow paths of heat and current between different nodes, thereby tracing the complete conduction path. After determining the conduction path, the detailed information of the start and end nodes is recorded, generating the conduction path tracing results.

[0019] Step S130: Based on the source tracing results of the conduction path, bind the conduction link tag of the corresponding conduction path to each continuous sampling record, and concatenate the continuous sampling records under the same conduction path in chronological order to generate a set of sampling records bound to the conduction path that is deeply associated with the continuous sampling records.

[0020] After obtaining the conduction path tracing results, a corresponding conduction link tag is assigned to each continuous sampling record corresponding to its conduction path. For example, if the conduction path corresponding to a continuous sampling record is determined to be path A, then a conduction link tag is assigned to path A for that record. Next, continuous sampling records under the same conduction path are concatenated in chronological order. Assuming that multiple continuous sampling records all correspond to path A, these records are connected sequentially according to their collection time. In the final set of sampling records bound to the conduction path, a deep association is established between the conduction path and the continuous sampling records. All sampling records under a certain conduction path can be quickly located through the conduction link tag.

[0021] Step S140: Arrange the sampling record set bound to the conduction path in chronological order of the entire operation cycle. Based on the timestamp of the same time node, establish an association index for continuous sampling records under different conduction paths to generate a cross-path time-series associated sampling record set.

[0022] In one implementation, step S140 may specifically include the following steps S141 to S147:

[0023] Step S141: Extract the time sequence markers of each continuous sampling record in the sampling record set bound to the conduction path, sort all the time sequence markers in chronological order, and generate an ordered time sequence marker sequence containing all time nodes.

[0024] Time sequence markers are the time information corresponding to each consecutive sampling record. By extracting these time sequence markers and sorting them in chronological order, an ordered sequence containing all time sequence nodes can be obtained. In practice, sorting algorithms, such as quicksort, can be used to sort the time sequence markers in the sampled record set.

[0025] Step S142: Match each continuous sampling record in the sampling record set bound to the conduction path with the corresponding time marker in the ordered time marker sequence to generate a time-ordered sampling record set.

[0026] For example, for a continuous sampling record whose acquisition time is the 5th minute, the time marker "5th minute" is found in the ordered time sequence, and the record is associated with this marker. By matching all sampling records one by one, they are rearranged in chronological order to generate a time-ordered set of sampling records.

[0027] Step S143: Extract all consecutive sampling records under the same time node from the temporally ordered sampling record set, and generate multi-path sampling record groups for the same time node based on their common timestamp.

[0028] In a time-ordered set of sampled records, there may be multiple sampled records from different conduction paths corresponding to the same time node. All consecutive sampled records at this same time node are extracted and grouped according to their common timestamp. For example, at the time node of the 8th minute, there are consecutive sampled records from paths A, B, and C. These records from these three paths are extracted and grouped together based on the common timestamp "8th minute," forming a multi-path sampled record group for the same time node. Such grouping helps in analyzing the interrelationships and influences of different conduction paths at the same time point.

[0029] Step S144: Add a unified timestamp identifier based on the time sequence node to the multi-path sampling record group of the same time sequence node to generate a sampling record group with timestamp identifier.

[0030] To more clearly identify each multipath sampling record group, a unified timestamp identifier based on the corresponding time sequence node is added to it. For example, for the multipath sampling record group at the 8th minute, a timestamp identifier of "8th minute" is added. In this way, each group in the generated sampling record group with timestamp identifier has clear time information.

[0031] Step S145: Arrange all sampling records with cross-path identifiers in chronological order to generate a set of sampling records with cross-path chronological association.

[0032] Cross-path identification refers to the grouping of sampling records across different conduction paths. By arranging these groups in chronological order, the sampling records across different conduction paths are correlated in the time dimension, generating a cross-path time-series correlated sampling record set. For example, by arranging the timestamped sampling record groups at different time nodes such as the 3rd minute, 5th minute, and 8th minute in sequence, a complete cross-path time-series correlated sampling record set is formed. This set can clearly show the temporal changes and interrelationships of the sampling records across different conduction paths.

[0033] Step S146: Perform a continuity check on the cross-path temporal correlation sampling record set, remove sampling record groups with cross-path identifiers that have temporal breaks, and generate a continuous and complete cross-path temporal correlation sampling record set.

[0034] The continuity check for cross-path association examines whether the sampled records under different conduction paths in a cross-path temporally associated sampled record set are temporally continuous. A time-disrupted sampled record group with a cross-path identifier is one that has missing or discontinuous data in its time series. For example, if a sampled record under a certain conduction path is missing between the 5th and 6th minutes, then the sampled record group containing the 5th and 6th minutes of that conduction path with the cross-path identifier may have a time-disrupted sequence.

[0035] During continuity checks, the presence of temporal breaks can be determined by comparing the grouping of adjacent timestamps. If a group is found to have a temporal break, it is removed from the set of sampled records associated with cross-path temporal correlation.

[0036] Step S147: Unify the cross-path association identifiers of the continuous and complete cross-path temporal association sampling record set, so that the format of all cross-path association identifiers is consistent, and generate a cross-path sampling record set with unified identifiers.

[0037] Cross-path association identifiers are used to identify the relationships between different transmission paths. Because inconsistencies in the format of cross-path association identifiers may occur during data acquisition and processing, standardization is necessary. For example, some cross-path association identifiers may use numeric encoding, while others may use alphabetic encoding.

[0038] During the standardization process, a unified identifier format rule is established, and all cross-path association identifiers in a continuous and complete cross-path temporal association sampling record set are converted according to this rule. For example, it is uniformly stipulated that cross-path association identifiers adopt a three-digit code, and the original identifiers of different formats are converted into three-digit codes.

[0039] Step S150: Unify and integrate the conduction link markings of the cross-path time-series associated sampling record set, so that the conduction link marking format of all continuous sampling records is consistent, and generate a multi-scale photoelectric and thermal time-series dataset anchored to the conduction link.

[0040] During data acquisition and labeling, the format of conduction link tags may differ due to variations in different sensor nodes or processing methods. For example, some conduction link tags may use letter codes, while others may use number codes. To standardize and integrate the data, a unified conduction link tag format rule is established, and all conduction link tags in the cross-path time-series associated sampling record set are converted according to this rule. For example, it is standardized that conduction link tags use an uppercase letter followed by three digits, such as "A001" and "B002". This converts the previously different conduction link tag formats into this unified format.

[0041] Step S160: Group and store the multi-scale photoelectric and thermal time series dataset of the anchored conduction link according to the conduction path dimension, and generate an independent storage group corresponding to each conduction path.

[0042] When grouping and storing multi-scale photoelectric and thermal time-series datasets anchored by conduction links according to the conduction path dimension, records belonging to the same conduction path are grouped together based on the conduction link label attached to each consecutive sampling record in the dataset. For example, if the dataset has three consecutive sampling records labeled "Path A", "Path B", and "Path A", the first two records are grouped into the group corresponding to "Path A", and the third record is grouped into the group corresponding to "Path B". This grouping method generates an independent storage group corresponding to each conduction path. The advantage of this approach is that it facilitates independent analysis and management of data from different conduction paths, improving the efficiency and accuracy of data processing.

[0043] Step S170: Perform unified labeling of the transmission path identifiers on the independent storage groups to ensure that the identifier of each independent storage group is consistent with the format of the transmission link mark, and generate a labeled group storage set.

[0044] The transmission path identifier is used to identify the transmission path corresponding to each independent storage packet. To ensure that the identifier of each independent storage packet is consistent with the format of the transmission link label, a unified labeling method is required. For example, in the previous example, the identifier formats for the independent storage packets corresponding to "Path A" and "Path B" might have been inconsistent. Now, following the unified transmission link labeling format rules, such as the uppercase letter plus three-digit encoding method mentioned earlier, the storage packet identifier corresponding to "Path A" is labeled as "A001," and the storage packet identifier corresponding to "Path B" is labeled as "B002."

[0045] Step S200: Based on the multi-scale photoelectric and thermal time-series dataset with anchored conduction links, conduction causal chains of photoelectric and thermal effects between different thermal runaway process stages are mined using causal convolutional networks. Pseudo-associative sampling records are removed through reverse causal verification logic to obtain a set of cross-stage coupling dominant causal chains.

[0046] Thermal runaway refers to the different stages a MiniLED display undergoes during operation, from normal operation to thermal runaway, such as the early, middle, and late stages. The photoelectric-thermal interaction causal chain describes the causal relationship between light, electricity, and heat in different thermal runaway stages, illustrating the transmission and influence of heat and current among different physical quantities.

[0047] In one implementation, step S200 may specifically include the following steps S210 to S270:

[0048] Step S210: Input the multi-scale photoelectric-thermal time-series dataset anchored by the conduction link into the causal convolutional network, mine the conduction causal chain of photoelectric-thermal interaction between different thermal runaway process stages, and generate an initial conduction causal chain sequence containing all potential related links.

[0049] After inputting a multi-scale photoelectric-thermal time-series dataset anchored by conduction links into a causal convolutional network, the network extracts and analyzes features from the data through convolution operations. The causal convolutional network can employ a one-dimensional convolutional neural network (1D-CNN) architecture. The input layer receives the dataset, and the convolutional kernels in the convolutional layers slide across the data, extracting features at different scales. For example, for data such as chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conductivity differences, the convolutional kernels can capture the relationships between them at different time scales. During the convolution operation, the network learns the potential causal relationships in the data. By continuously adjusting the weights and biases of the convolutional kernels, the network can better fit the causal patterns in the data. For example, assuming that during a certain stage of thermal runaway, an increase in chip junction temperature leads to a decrease in electro-optical conversion efficiency, the network learns this pattern of change in the data to uncover the causal relationship between the two. Finally, the network outputs an initial conduction causal chain sequence containing all potentially related links. Each conduction causal chain in this sequence may represent a different photoelectric-thermal causal relationship, but some spurious links may exist, requiring further verification and processing.

[0050] Step S220: Divide the initial conduction causal chain sequence into thermal runaway process stages, and assign each conduction causal chain to the corresponding thermal runaway process stage according to the time sequence node range it covers, generating a multi-stage conduction causal chain sub-sequence.

[0051] In one implementation, step S220 may specifically include the following steps S221 to S227:

[0052] Step S221: Extract the start and end time nodes of each causal chain in the initial causal chain sequence, record the time range covered by each causal chain, and generate a causal chain time boundary set containing the time boundary information of each causal chain.

[0053] In this step, each causal chain in the initial causal chain sequence is analyzed in detail. By processing the timestamp information of the data in the causal chain, its start and end time nodes are extracted. For example, for a given causal chain, by examining the timestamps of its sampled records, the earliest record time is found to be 5 minutes and the latest record time is 15 minutes. Therefore, the start time node of this causal chain is 5 minutes, and the end time node is 15 minutes, covering a time range from 5 minutes to 15 minutes.

[0054] Record the temporal boundary information (starting temporal node and ending temporal node) of each causal chain to form a causal chain temporal boundary set containing the temporal boundary information of each causal chain.

[0055] Step S222: Extract the time sequence node range of each stage of the thermal runaway process within the entire operating cycle, record the start and end time points of each thermal runaway process stage, and generate a set of stage time sequence ranges containing the complete time range of each thermal runaway process stage.

[0056] The entire operating cycle refers to the entire process of a MiniLED display screen from startup to shutdown. By analyzing the historical operating data and thermal runaway characteristics of the display screen, the timing range of each stage of the thermal runaway process is determined. For example, based on previous experiments and observations, it was found that the early stage of thermal runaway extends from the display screen startup to the 10th minute, the middle stage from the 10th minute to the 30th minute, and the late stage from the 30th minute to the display screen shutdown. Therefore, the start time of the early stage is the display screen startup time, and the end time is the 10th minute; the start time of the middle stage is the 10th minute, and the end time is the 30th minute; the start time of the late stage is the 30th minute, and the end time is the display screen shutdown time. By recording the start and end times of each thermal runaway process stage, a set of stage timing ranges containing the complete time range of each thermal runaway process stage is generated.

[0057] Step S223: Match the temporal boundary of each conducted causal chain in the causal chain temporal boundary set with the temporal range of the corresponding thermal runaway process stage in the stage temporal range set to determine the thermal runaway process stage to which each conducted causal chain belongs, and generate the matching result between the conducted causal chain and the thermal runaway process stage.

[0058] When matching the causal chain temporal boundary set with the stage temporal range set, for each propagating causal chain in the causal chain temporal boundary set, its start and end time nodes are compared with the start and end time points of each thermal runaway process stage in the stage temporal range set. If the start and end time points of a propagating causal chain fall entirely within the time range of a thermal runaway process stage, then the propagating causal chain is determined to belong to that thermal runaway process stage.

[0059] For example, suppose a causal chain starts at minute 12 and ends at minute 25, while the middle stage in the set of stage time ranges starts at minute 10 and ends at minute 30. Then this causal chain belongs to the middle stage. By performing this matching operation on all causal chains in the causal chain time boundary set, matching results between causal chains and thermal runaway process stages are generated.

[0060] Step S224: Based on the matching results of the conduction causal chain and the thermal runaway process stage, the conduction causal chains belonging to the same thermal runaway process stage are grouped to generate a stage causal chain grouping set corresponding to each thermal runaway process stage.

[0061] In one implementation, step S224 may specifically include the following steps S2241 to S2247:

[0062] Step S2241: Extract all the conduction causal chains corresponding to each thermal runaway process stage from the matching results of the conduction causal chains and thermal runaway process stages, summarize the conduction paths, time boundaries and other information of all conduction causal chains, and generate a candidate set of causal chains for each thermal runaway process stage.

[0063] In the matching results between the conduction causal chains and the thermal runaway process stages, each thermal runaway process stage corresponds to several conduction causal chains. For each thermal runaway process stage, all corresponding conduction causal chains are extracted. Then, information such as the conduction path and timing boundaries of these conduction causal chains are summarized. The conduction path is the specific path through which heat, current, etc., are transferred between the chip-substrate-package coupling nodes, and the timing boundary is the start and end point of the conduction causal chain in time.

[0064] For example, for the intermediate stage, assuming the matching results show that there are five causal chains belonging to this stage, the transmission path information (such as path A, path B, etc.) and time boundary information (such as 10 minutes to 20 minutes, 12 minutes to 25 minutes, etc.) of these five causal chains are summarized to form a candidate set of causal chains for the intermediate stage.

[0065] Step S2242: Compare the similarity of the conduction paths of the conduction causal chains in the candidate set of causal chains for each thermal runaway process stage, and cluster the conduction causal chains with similar conduction path structures and the same active nodes to generate a cluster subset of causal chains within the thermal runaway process stage.

[0066] Similarity comparison of conduction paths is achieved by comparing the conduction path structure and active nodes of the conduction causal chain. The conduction path structure is the specific way and order in which heat and current are transferred between the chip-substrate-package coupling nodes, while the active nodes are the nodes that play a key role in the conduction path, such as the heat dissipation pins of the chip and the heat conduction areas on the substrate.

[0067] When comparing the similarity of conduction paths in the candidate set of causal chains for each stage of thermal runaway, a graph matching algorithm can be used. First, the conduction path of each causal chain is represented as a graph structure, where nodes represent coupling nodes and edges represent the conduction relationships between nodes. Then, the similarity between graphs is calculated to determine whether the conduction paths are similar. If two causal chains have similar conduction path structures and the same active nodes, they are clustered into a subset.

[0068] For example, in the candidate set of causal chains in the intermediate stage, there are two conduction causal chains whose conduction paths both start from a certain heat dissipation pin of the chip, pass through the same heat conduction area on the substrate, and finally reach a certain heat dissipation part of the package. The nodes of action are both this heat dissipation pin and the heat conduction area. Then these two conduction causal chains are clustered into a subset.

[0069] Step S2243: Integrate the conduction correlations of the causal chain cluster subsets within each thermal runaway process stage, connect the conduction causal chains with direct conduction correlations in the causal chain cluster subsets to form a complete conduction link in the causal chain cluster subsets, and generate the integrated stage causal chain subset.

[0070] A direct conduction correlation is a direct relationship between two conduction causal chains that involves the transfer of heat or current. For example, one conduction causal chain describes the process of heat generated by the chip being transferred to the substrate, while another conduction causal chain describes the process of heat on the substrate being transferred to the package. These two conduction causal chains have a direct conduction correlation.

[0071] When integrating the causal chain clusters within each stage of thermal runaway, the relationships between them can be determined by analyzing the transmission paths and timing information of the causal chains. If two causal chains are found to have a direct transmission relationship, they are connected in series according to their transmission order. For example, the two causal chains described above—one from chip to substrate and the other from substrate to package—are connected in series to form a complete transmission link from chip to package. By integrating all causal chain clusters in this way, an integrated set of stage causal chains is generated.

[0072] Step S2244: Group and integrate all the integrated stage causal chain subsets within the same thermal runaway process stage, and merge all the integrated stage causal chain subsets within the thermal runaway process stage into a complete set to generate the stage causal chain grouping set corresponding to the thermal runaway process stage.

[0073] After obtaining all integrated subsets of stage causal chains within the same thermal runaway process phase, these subsets are grouped and integrated. For example, in the intermediate phase, multiple causal chain clusters may be integrated to form multiple integrated subsets of stage causal chains. These subsets are then merged into a complete set, forming the stage causal chain grouping set corresponding to the intermediate phase.

[0074] Step S2245: Verify the grouping results of the causal chain grouping set for each stage, check whether the grouped causal chains all belong to the corresponding thermal runaway process stage, remove erroneous grouped links that do not belong to the thermal runaway process stage, and generate grouping verification results.

[0075] The verification of the grouping results is to ensure that all the propagational causal chains in the stage causal chain grouping set are correctly assigned to the corresponding thermal runaway process stage. During the verification process, the temporal boundary information of each propagational causal chain is checked again and compared with the time range of the corresponding thermal runaway process stage. If the start or end time of a propagational causal chain is found to be outside the time range of the corresponding thermal runaway process stage, then that propagational causal chain is an incorrectly grouped link and is removed from the stage causal chain grouping set.

[0076] Step S2246: Based on the grouping verification results, adjust the grouping set of stage causal chains to ensure that all conduction causal chains in the set belong to the corresponding thermal runaway process stage, and generate the final grouping set of stage causal chains.

[0077] Based on the grouping verification results, the grouping set of stage causal chains is adjusted accordingly. If an incorrectly grouped link is found and removed during the verification process, the content of the stage causal chain grouping set will change. The set after removing the incorrectly grouped link is updated to ensure that all propagation causal chains within the set correctly belong to the corresponding thermal runaway process stage. For example, in the intermediate stage, after grouping verification, an incorrectly grouped link is removed, and the remaining propagation causal chains are recombined to form a new set. This new set is the final stage causal chain grouping set.

[0078] Step S2247: Add grouping identifiers to the final stage causal chain grouping set. Add the corresponding grouping identifier for the thermal runaway process stage to each stage causal chain grouping set to generate a stage causal chain grouping set with grouping identifiers.

[0079] Grouping identifiers are used to identify the thermal runaway process stage corresponding to each stage causal chain grouping set. To more clearly manage and distinguish stage causal chain grouping sets of different thermal runaway process stages, a grouping identifier corresponding to the thermal runaway process stage is added to each set. For example, for stage causal chain grouping sets of early stages, a grouping identifier of "early stage" is added; for sets of intermediate stages, a grouping identifier of "intermediate stage" is added; and for sets of late stages, a grouping identifier of "late stage" is added.

[0080] Step S225: Perform temporal continuity integration of the conduction causal chains in the grouping set of each stage causal chain, connect the temporally continuous conduction causal chains within the same thermal runaway process stage to form a complete conduction link within the thermal runaway process stage, and generate a continuous stage causal chain subsequence.

[0081] Temporal continuity integration involves connecting and integrating temporally continuous causal chains within the same thermal runaway process phase. Temporal continuity means that the end time of one causal chain is adjacent to or close to the start time of another causal chain. For example, if one causal chain ends at minute 20 and another begins at minute 21, these two causal chains are temporally continuous.

[0082] When performing temporal continuity integration on the causal chain grouping set for each stage, the temporal boundary information of each causal chain in the set is first analyzed. Then, temporally continuous causal chains are identified and connected in chronological order. For example, in the mid-stage causal chain grouping set, there are three causal chains with temporal boundaries of 10-20 minutes, 20-25 minutes, and 25-30 minutes, respectively. These three causal chains are temporally continuous, and connecting them forms a complete transmission link from 10 minutes to 30 minutes.

[0083] Step S226: Bind multiple consecutive stage causal chain sequences to thermal runaway process stage identifiers, add a unique identifier for each consecutive stage causal chain sequence corresponding to the thermal runaway process stage, and generate multiple stage conduction causal chain sequences.

[0084] To more clearly identify the thermal runaway process stage corresponding to each consecutive causal chain subsequence, multiple consecutive causal chain subsequences are bound with thermal runaway process stage identifiers. For example, for consecutive causal chain subsequences in the early stage, a unique identifier for "early stage" is added; for subsequences in the middle stage, a unique identifier for "middle stage" is added; and for subsequences in the late stage, a unique identifier for "late stage" is added.

[0085] Step S227: Verify the stage identifiers of the multi-stage causal chain subsequences, ensuring that the identifier of each stage causal chain subsequence is completely consistent with the corresponding thermal runaway process stage, and generate a set of verified stage causal chain subsequences.

[0086] The verification of stage identifiers is to ensure the consistency between the identifiers of each stage's causal chain subsequence and the corresponding thermal runaway process stage. During the verification process, the temporal boundary information and identifier information of each subsequence are checked again, and the temporal boundaries are compared with the time range of the corresponding thermal runaway process stage to ensure that the identifiers match the actual stages.

[0087] For example, if a subsequence is identified as "mid-stage," but inspection reveals that its timeframe actually belongs to the early stage, then the identification of that subsequence needs to be corrected. By performing such stage identification verification and correction operations on all multi-stage causal chain subsequences, a set of verified stage causal chain subsequences is generated.

[0088] Step S230: Verify the causal direction of each stage of the causal chain subsequence, check the rationality of the causal direction of each causal chain through reverse causal verification logic, and generate a causal direction verification result containing the verification results of each causal chain.

[0089] When validating the causal direction of each stage's causal chain subsequence, the variables included in each causal chain are analyzed. For example, for a causal chain describing the decrease in electro-optical conversion efficiency due to increased chip junction temperature, the Granger causality test is used to verify whether there is a significant difference in the predictive ability of electro-optical conversion efficiency with and without the chip junction temperature variable. If the predictive ability of electro-optical conversion efficiency is significantly improved when the chip junction temperature variable is included, then the increase in chip junction temperature can be considered the cause of the decrease in electro-optical conversion efficiency, and the causal direction is reasonable; conversely, if there is no significant difference, then the causal direction may be unreasonable. By performing such causal direction validation on all causal chains in each stage's causal chain subsequence, a causal direction validation result containing the validation results of each causal chain is generated.

[0090] Step S240: Based on the causal direction verification results, remove the pseudo-correlation links with unreasonable causal directions in the stage-transmission causal chain sub-sequence and generate the stage-dominant causal chain sub-sequence.

[0091] After obtaining the causal direction verification results, the stage-directed causal chain subsequences are processed based on these results. If the causal direction verification result of a certain causal chain is unreasonable, then that causal chain is a pseudo-association link and is removed from the stage-directed causal chain subsequence. For example, in a certain stage-directed causal chain subsequence, after causal direction verification, it is found that the causal relationship direction of one causal chain does not conform to reality, that is, the causal relationship it describes does not exist in reality, and this causal chain is removed from the sequence. By performing this removal operation on all stage-directed causal chain subsequences, a stage-dominant causal chain subsequence is generated.

[0092] Step S250: Perform cross-stage correlation integration on the multi-stage dominant causal chain subsequences, and connect the conduction causal chains that have conduction correlation between adjacent thermal runaway process stages to generate a cross-stage coupled dominant causal chain set.

[0093] Cross-stage correlation integration involves connecting and integrating the causal chains of conduction relationships between different stages of thermal runaway. Adjacent thermal runaway stages are those that are temporally adjacent, such as the early and middle stages, or the middle and late stages. Conduction relationships refer to the transfer of heat, current, etc., between the causal chains of conduction in different stages.

[0094] When performing cross-stage correlation integration, the dominant causal chain sequences of adjacent thermal runaway process stages are analyzed. For example, the sequences of the early and middle stages are analyzed to identify conduction causal chains with conductive correlations. If an early stage conduction causal chain describes the initial rise in chip junction temperature, and a middle stage conduction causal chain describes the further impact of the rise in chip junction temperature on electro-optical conversion efficiency, and there is a heat transfer correlation between these two conduction causal chains, then they are connected in series.

[0095] Step S260: Sort the set of causal chains that dominate cross-stage coupling by causal relationships, and arrange them from high to low according to the tightness of causal relationships, to generate a set of causal chains sorted by the tightness of relationships.

[0096] The strength of a causal relationship can be measured by calculating the correlation between variables in the causal chain. For example, the Pearson correlation coefficient can be used to calculate the correlation between chip junction temperature and electro-optical conversion efficiency; the closer the absolute value of the correlation coefficient is to 1, the stronger the causal relationship between the two.

[0097] When ranking the causal relationships of a set of dominant causal chains across stages, the correlation between variables in each transmission causal chain in the set is first calculated. Then, based on the magnitude of the correlation, the transmission causal chains are arranged from high to low in terms of the tightness of the causal relationship.

[0098] Step S270: Standardize the labeling of the causal chain set sorted by the degree of correlation to ensure that the labeling format of the correlation strength of each causal chain is consistent, and generate a labeled causal chain set.

[0099] Association strength refers to the degree of causal relationship between variables in a causal chain. The tightness of causal association in the causal chain was previously determined by methods such as calculating the Pearson correlation coefficient, and this tightness can be used as a measure of association strength. Each causal chain in the set of causal chains sorted by association strength is labeled with its association strength using a uniform labeling format, ensuring consistency in the labeling format for association strength across all causal chains.

[0100] Step S300: Using the set of cross-stage coupling dominant causal chains as the core input, continuous sampling records of chip manufacturing discreteness and environmental thermal disturbances are introduced for bidirectional mutual feedback calibration. The causal chain deviation transmission identification logic is used to remove the transmission deviation causal chains generated during the calibration process, and the coupled transmission causal network after mutual feedback correction is obtained.

[0101] Chip manufacturing variability arises from factors such as manufacturing processes, resulting in differences in physical properties between different chips. For example, parameters like resistance and capacitance may vary, affecting the chip's heat generation and photoelectric conversion performance. Environmental thermal disturbances refer to changes in temperature, humidity, and other factors in the environment in which the MiniLED display is located. These changes can interfere with the display's heat conduction and heat dissipation.

[0102] Two-way feedback calibration involves continuously sampling records of chip manufacturing discreteness and environmental thermal disturbances and mutually feeding back and calibrating them with the set of dominant causal chains across stages. On one hand, information on chip manufacturing discreteness and environmental thermal disturbances is used to correct the conduction causal chains in the set of dominant causal chains across stages; on the other hand, the impact of chip manufacturing discreteness and environmental thermal disturbances is evaluated and adjusted based on the conduction relationships in the set of dominant causal chains across stages.

[0103] In one implementation, step S300 may specifically include the following steps S310 to S370:

[0104] Step S310: Using the set of cross-stage coupling dominant causal chains as the core input, introduce continuous sampling records of chip manufacturing discreteness, correct the correlation or state parameters of relevant transmission path nodes in the transmission causal chain according to the manufacturing discrete information, perform mutual feedback calibration in the first direction, and generate the first calibrated causal chain set.

[0105] After taking the set of cross-stage coupling-dominant causal chains as the core input, continuous sampling records of chip manufacturing discreteness are introduced. These continuous sampling records contain information on the differences in physical characteristics generated by different chips during the manufacturing process, such as discrete values ​​of parameters like resistance and capacitance.

[0106] Based on manufacturing discrete information, the correlations or state parameters of relevant conduction path nodes in the conduction causal chain are corrected. For example, for a conduction path node related to a certain chip in the conduction causal chain, if the resistance of that chip is higher than that of a normal chip due to manufacturing discreteness, then the state parameters of that node, such as thermal conductivity and current conduction capability, need to be adjusted accordingly. At the same time, the correlations between the nodes before and after that node in the conduction path may also need to be changed to reflect the impact of resistance changes on the entire conduction path.

[0107] In one implementation, step S310 may specifically include the following steps S311 to S316:

[0108] Step S311: Extract the transmission path node information of each transmission causal chain in the cross-stage coupling dominant causal chain set, record the position, target and relationship of each transmission path node, and generate a transmission path node set of the transmission causal chain.

[0109] In this step, each conductive causal chain in the set of dominant causal chains for cross-stage coupling is analyzed in detail. Conductive path nodes are key nodes in the transfer of heat, current, etc., within the conductive causal chain, such as chip pins or heat conduction areas on the substrate. By analyzing the conductive paths of the conductive causal chains, the location information of each conductive path node is extracted, i.e., the specific location of the node in the chip, substrate, or package; the information of the object of influence, i.e., the object of influence of the node on other surrounding nodes, such as the heat conduction effect of a chip pin on an adjacent substrate area; and the information of relationships, i.e., the connection method and mutual influence relationship between the node and other conductive path nodes, such as the direction of heat conduction and the intensity of current conduction.

[0110] Step S312: Extract manufacturing discrete information from the continuous sampling records of chip manufacturing discreteness, record the source, scope of influence and mode of action of each manufacturing discrete information, associate each manufacturing discrete information with the conduction path node of the corresponding chip-substrate-package coupling node, and generate the association result between discrete information and node.

[0111] The continuous sampling records of chip manufacturing discreteness contain various discrete information generated during the manufacturing process of different chips, such as discrete values ​​of parameters like resistance and capacitance. First, this manufacturing discrete information is extracted from the continuous sampling records, recording the source of each piece of discrete information (i.e., which specific chip it originates from); the scope of influence (i.e., which regions or nodes the discrete information affects, such as the discrete resistance value of a chip affecting the chip itself and the substrate area connected to it); and the mode of action (i.e., how the discrete information affects other nodes, such as how changes in resistance affect current conduction and heat generation). Then, each piece of manufacturing discrete information is associated with the corresponding conduction path nodes of the chip-substrate-package coupling node. For example, for the discrete resistance information of a certain chip, the nodes in the conduction path related to that chip are found, and the discrete resistance information is associated with these nodes. Through this association operation, the association results between discrete information and nodes are generated, clarifying the correspondence between each piece of manufacturing discrete information and the conduction path nodes.

[0112] Step S313: Based on the association results between discrete information and nodes, use the manufacturing discrete information to correct the association information or state parameters of the corresponding transmission path nodes, and generate a corrected causal chain node set.

[0113] In one implementation, step S313 may specifically include the following steps S3131 to S3138:

[0114] Step S3131: Extract the manufacturing discrete information corresponding to each transmission path node in the association results of discrete information and nodes, extract and summarize the source, scope of influence and mode of action of each manufacturing discrete information, and generate a set of correspondences between nodes and discrete information containing complete attributes of manufacturing discrete information.

[0115] In the association results between discrete information and nodes, each conduction path node corresponds to relevant manufacturing discrete information. A detailed analysis of these association results extracts the manufacturing discrete information corresponding to each conduction path node. For example, a conduction path node located at a specific position on a chip might correspond to the manufacturing discrete information of the chip's resistance value. For each piece of manufacturing discrete information, its source is extracted in detail, clarifying which specific manufacturing process or batch of chips caused it; its scope of influence is determined, such as which circuit areas inside the chip are affected by the discrete resistance value, and to what extent it affects the connected substrate; its mode of action is analyzed, such as how changes in resistance value affect the heat conduction process by altering the current. These complete attributes regarding the source, scope of influence, and mode of action of the manufacturing discrete information are summarized, establishing a complete description of the corresponding manufacturing discrete information for each conduction path node, ultimately generating a set of correspondences between nodes and discrete information containing complete attributes of the manufacturing discrete information.

[0116] Step S3132: Based on the set of correspondences between nodes and discrete information, each piece of manufacturing discrete information is added to the attribute set of the corresponding transmission path node as an influencing factor, and the node's association information or state parameters are updated accordingly to generate updated node association information.

[0117] After obtaining the set of correspondences between nodes and discrete information, each piece of manufacturing discrete information is considered an important influencing factor. For each conduction path node, based on the correspondence set, the relevant manufacturing discrete information is added to the node's attribute set. For example, if a conduction path node corresponds to the manufacturing discrete information of chip capacitance value, this capacitance discrete information is added to the node's attribute set. Then, based on this newly added manufacturing discrete information, the node's association information or state parameters are updated. Taking capacitance discrete information as an example, changes in capacitance affect the circuit's charging and discharging time, thus affecting current conduction and heat generation. Therefore, it is necessary to recalculate the node's current conduction capability, heat generation rate, and other state parameters, as well as its association information such as the association strength with other nodes, based on the capacitance discrete information. Through this operation, updated node association information is generated, enabling the node information to more accurately reflect the impact of chip manufacturing discreteness.

[0118] Step S3133: Update the integrated node association information to the corresponding transmission path node of the transmission causal chain, replace the original node association information, ensure that the node information accurately reflects the impact of manufacturing discrete information, and generate the updated transmission path node.

[0119] After obtaining the updated node association information, this integrated information is applied to the corresponding conduction path nodes in the conduction causal chain. Specifically, for each conduction path node in the conduction causal chain, its original association information is replaced with the updated association information. For example, in a conduction causal chain describing heat conduction from a chip to a substrate, the original association information of a certain conduction path node was based on standard chip parameters. Now, after updating the association information of this node according to the discrete chip manufacturing information, the new association information replaces the original information. The purpose of this is to ensure that the information of the conduction path nodes accurately reflects the impact of the discrete manufacturing information. Through this replacement operation, updated conduction path nodes are generated, enabling the conduction causal chain to more realistically simulate the photoelectric and thermal conduction process under the influence of chip manufacturing discreteness.

[0120] Step S3134: Group all updated transmission path nodes according to their respective transmission causal chains, integrate the updated transmission path nodes under the same transmission causal chain into the same set, and generate the updated node set corresponding to each transmission causal chain.

[0121] After obtaining all updated conduction path nodes, they are grouped according to their respective conduction causal chains. Since different conduction causal chains describe different photoelectric and thermal conduction processes, integrating updated conduction path nodes under the same conduction causal chain facilitates overall analysis and processing of each conduction causal chain. For example, for conduction causal chains describing internal chip thermal conduction and those describing chip-to-substrate thermal conduction, the updated conduction path nodes belonging to these two chains are each integrated into a separate set. Through this grouping and integration operation, an updated node set corresponding to each conduction causal chain is generated.

[0122] Step S3135: Perform a node integrity check on the update node set corresponding to each causal chain, remove nodes with missing information or incorrect associations, ensure the integrity and accuracy of the update node set, and generate a complete update node set.

[0123] Within the set of updated nodes corresponding to each causal chain, there may be some nodes with missing information or incorrect associations. Missing information may be due to the incorrect recording of certain manufacturing discrete information during data acquisition or processing, resulting in incomplete attribute information for some nodes. Incorrect associations may occur when associating manufacturing discrete information with nodes or updating node association information, causing the relationships between nodes to be inconsistent with reality. A node integrity check is performed on each set of updated nodes, using a series of check rules to determine whether nodes are complete and accurate. For example, checking whether the node's attribute information contains all necessary parameters and whether the relationships between nodes and other nodes conform to physical laws. If a node is found to have missing information or incorrect associations, it is removed from the set of updated nodes.

[0124] Step S3136: Summarize the complete set of update nodes corresponding to all causal chains to generate the injected causal chain node set.

[0125] After obtaining the complete set of updated nodes for each causal chain, these sets are aggregated. All complete sets of updated nodes for all causal chains are merged together to form a large set containing all updated path nodes, i.e., the injected causal chain node set. This set integrates the updated node information of all causal chains after considering the discrete information of chip manufacturing, providing a comprehensive data source for subsequent unified processing and analysis of the entire causal chain set.

[0126] Step S3137: Verify the injection result of the injected causal chain node set, check whether the node information is completely consistent with the manufacturing discrete information, and generate a causal chain node set with verification mark.

[0127] Verifying the injected causal chain node set ensures that the node information accurately reflects the impact of the manufacturing discrete information. This involves checking whether the node's attribute and association information is completely consistent with the previously used manufacturing discrete information. For example, it checks whether the node's resistance, capacitance, and other attributes have been correctly updated according to the corresponding manufacturing discrete information, and whether the relationships between nodes reasonably reflect the impact of the manufacturing discrete information on the conduction process. Various verification methods can be used, such as comparative verification, comparing the updated node information with the original manufacturing discrete information to check for discrepancies; and simulation verification, using a physical model to simulate the updated node information and checking whether the simulation results meet expectations. If the node information is consistent with the manufacturing discrete information, a verification pass flag is added to the node; if inconsistencies exist, the node information is corrected until verification is successful.

[0128] Step S3138: Add injection identifiers to the set of causal chain nodes with verification identifiers, add an injection-specific identifier to each updated propagation path node, and generate a set of causal chain nodes with injection identifiers.

[0129] To facilitate the management and tracking of transmission path nodes after correction using discrete chip manufacturing information, an injection identifier is added to the set of causal chain nodes with verification tags. Each updated transmission path node is given an injection-specific identifier, which can be a specific code or mark used to uniquely identify that the node has been corrected by injecting discrete chip manufacturing information. For example, the format "MDI-[node number]" can be used as the injection-specific identifier, where "MDI" represents the chip manufacturing discrete information injection, and "[node number]" is the unique number of the node in the entire causal chain system. By adding injection-specific identifiers, a set of causal chain nodes with injection identifiers is generated.

[0130] Step S314: Reconnect the transmission paths of the injected causal chain node set, and connect the updated transmission path nodes according to the temporal order of the original causal chain to ensure the temporal continuity of the transmission path, and generate the reconnected causal chain set.

[0131] After obtaining the set of causal chain nodes with injection identifiers, the conduction paths need to be reconnected. This is because the information of the conduction path nodes changes after the chip manufacturing discrete information is corrected, which may affect the connection and order of the conduction paths. The updated conduction path nodes are connected according to the original temporal order of the conduction causal chain. For example, in a conduction causal chain describing the chip's heat conduction process, the original temporal order is from the chip's heat source node to the intermediate heat dissipation channel node, and then to the heat dissipation area node on the substrate. After updating the node information, the updated corresponding nodes are reconnected according to this temporal order. This reconnection operation ensures the temporal continuity of the conduction paths, allowing heat, current, and other substances to be transferred in the correct order and time sequence during conduction.

[0132] Step S315: Perform first-direction mutual feedback calibration on the reconnected causal chain set, adjust the correlation strength of the transmission path, so that the transmission path of the causal chain is consistent with the influence of the manufacturing discrete information, and generate the first-calibrated causal chain set.

[0133] The primary purpose of performing first-direction mutual feedback calibration on the reconnected causal chain set is to adjust the correlation strength of the conduction paths to accommodate the influence of discrete information from chip manufacturing. The correlation strength of a conduction path reflects the degree of interaction between nodes in different conduction paths, such as the intensity of heat conduction and current conduction. Since discrete information from chip manufacturing alters the properties and states of nodes, thus affecting the interactions between them, the correlation strength of the conduction paths needs adjustment. For example, when the resistance of a chip increases due to manufacturing discreteness, the current decreases, which in turn affects the heat conduction and current conduction strengths of the conduction paths associated with that chip. Therefore, the correlation strength of these conduction paths needs to be reduced accordingly. By establishing a physical model and mathematical algorithms, the correlation strength of the conduction paths is calculated and adjusted based on the discrete information from manufacturing. Specifically, calculation methods based on physical laws, such as Ohm's law and the law of thermal conduction, can be used, combined with the influence of discrete information from manufacturing on node properties, to recalculate the correlation strength of the conduction paths. After this adjustment, the conduction paths of the causal chains are aligned with the influence of discrete information from manufacturing, generating a first-calibrated causal chain set.

[0134] Step S316: Associate the calibration results of the first calibrated causal chain set with the calibration results, add a manufacturing discrete calibration-specific identifier to each first calibrated causal chain, generate a causal chain set with calibration tags, verify the calibration tags of the causal chain set with calibration tags, ensure that each calibration tag is consistent with the source of the corresponding manufacturing discrete information, and generate a verified calibration tag causal chain set.

[0135] To facilitate the management and traceability of the causal chains after the first-direction feedback calibration, the calibration results are associated with the set of causal chains after the first calibration. A manufacturing discrete calibration-specific identifier is added to each causal chain after the first calibration. This identifier can use a specific encoding format, such as "MDC-[causal chain number]", where "MDC" indicates manufacturing discrete calibration, and "[causal chain number]" is the unique number of that causal chain within the entire causal chain set. By adding the manufacturing discrete calibration-specific identifier, a set of causal chains with calibration tags is generated.

[0136] Next, the calibration identifiers of the causal chain set with calibration marks are verified. It is ensured that each calibration identifier is consistent with the source of the corresponding manufacturing discrete information; that is, it is checked whether the calibration process represented by the calibration identifier of each causal chain is indeed based on the chip manufacturing discrete information involved in that causal chain. Verification can be performed by reviewing records and data from the calibration process and comparing the correspondence between calibration identifiers and manufacturing discrete information. If an inconsistency is found between the source of the calibration identifier and the manufacturing discrete information, the calibration identifier is corrected until verification is successful.

[0137] Step S320: Using the first calibrated causal chain set as input, introduce continuous sampling records of environmental thermal disturbances, and correct the correlation or state parameters of relevant conduction path nodes in the conduction causal chain again based on the thermal disturbance information, perform mutual feedback calibration in the second direction, and generate the second calibrated causal chain set.

[0138] Ambient thermal disturbances refer to changes in factors such as temperature, humidity, and airflow in the environment in which a MiniLED display is located. These factors significantly affect the heat conduction and dissipation of the display. The first calibrated causal chain set is used as input to incorporate continuous sampling records of the ambient thermal disturbances. These continuous sampling records contain detailed information about the ambient thermal disturbances, such as changes in ambient temperature, humidity fluctuations, and airflow speed at different times.

[0139] The correlations and state parameters of relevant conduction path nodes in the conduction causal chain are further corrected based on thermal disturbance information. For example, when the ambient temperature rises, it affects the chip's heat dissipation efficiency, thus requiring adjustment of state parameters such as the thermal conductivity coefficient of the conduction path nodes related to chip heat dissipation in the conduction causal chain. Simultaneously, changes in ambient airflow may alter the heat conduction path between the substrate and the package, necessitating corresponding adjustments to the correlations between nodes. During the second-direction mutual feedback calibration, the conduction causal chain is dynamically adjusted by combining the heat transfer physics model and environmental thermal disturbance data. For instance, the degree of influence of environmental thermal disturbances on conduction path nodes is calculated using the heat conduction equation and fluid dynamics model, and the state parameters and correlations of the nodes are corrected accordingly. Through this further correction operation, a second-calibrated causal chain set is generated, enabling the conduction causal chain to more accurately reflect the impact of environmental thermal disturbances on the photoelectric thermal conduction process of the MiniLED display.

[0140] Step S330: Perform deviation propagation identification on the second calibrated causal chain set, detect the propagation deviation causal chains generated during the calibration process through the causal chain deviation propagation identification logic, and generate a deviation propagation causal chain set containing all propagation deviation causal chains.

[0141] The identification of conduction deviations in the second-calibrated causal chain set can be achieved through various methods, such as verification based on physical models to check whether the heat conduction and current conduction processes in the conduction causal chain conform to the laws of heat transfer and circuit laws; and logical consistency checks to examine whether the relationships between nodes in the causal chain are reasonable and whether there are any inverted or discontinuous causal relationships. For example, if heat flows from a low-temperature region to a high-temperature region in a conduction causal chain, this clearly does not conform to the basic principles of heat transfer, and this causal chain may be a conduction deviation causal chain. By performing such identification and detection on each conduction causal chain in the second-calibrated causal chain set, all causal chains identified as conduction deviation causal chains are collected to generate a deviation conduction causal chain set containing all conduction deviation causal chains. This set provides a clear target for subsequent removal of deviation causal chains and improvement of the accuracy of conduction causal chains.

[0142] Step S340: Compare the second calibrated causal chain set with the biased conduction causal chain set, remove the conduction biased causal chains that belong to the biased conduction causal chain set from the second calibrated causal chain set, and retain only the calibrated and valid conduction causal chains to generate the corrected causal chain set.

[0143] After obtaining the set of biased causal chains, the second calibrated causal chain set is compared with this set. For each causal chain in the second calibrated causal chain set, it is checked whether it exists in the set of biased causal chains. If it exists, it indicates that the causal chain is a biased causal chain and is removed from the second calibrated causal chain set; if it does not exist, it indicates that the causal chain is a calibrated and valid causal chain and is retained. Through this comparison and removal operation, a corrected causal chain set is generated. This set contains only calibrated and valid causal chains, excluding unreasonable causal chains generated during the environmental thermal disturbance calibration process, so that the causal chains can more accurately reflect the real photoelectric and thermal conduction process of the MiniLED display under the conditions of chip manufacturing discreteness and environmental thermal disturbance.

[0144] Step S350: Perform cross-stage association integration on the modified causal chain set, connect all the modified transmission causal chains in series according to the natural association of the transmission path, form a complete transmission network covering the entire operation cycle, and generate a mutually feed-back modified coupled transmission causal network.

[0145] The corrected causal chain set includes valid conducted causal chains that have been calibrated twice, once for chip manufacturing discreteness and once for environmental thermal disturbances, and once biased causal chains have been eliminated. These conducted causal chains are then integrated across stages and connected in series according to the natural correlation of the conduction paths. Different conducted causal chains may describe the photoelectric and thermal conduction processes of a MiniLED display at different stages of thermal runaway or in different physical regions. By analyzing the conduction path relationships between them, naturally correlated conducted causal chains are connected. For example, a conducted causal chain describing internal chip thermal conduction and a conducted causal chain describing chip-to-substrate thermal conduction have a natural correlation in heat transfer; these two causal chains are then connected in series.

[0146] By integrating all corrected conduction causal chains across different stages, a complete conduction network covering the entire operating cycle is formed. This network encompasses the transmission and interaction of light, electricity, and heat between various components such as the chip, substrate, and packaging throughout the entire process from the MiniLED display's startup to shutdown. A mutually feedback-corrected coupled conduction causal network is generated, which comprehensively considers the impact of chip manufacturing discreteness and environmental thermal disturbances, and eliminates unreasonable conduction causal chains, accurately reflecting the true photoelectric and thermal conduction process of the MiniLED display.

[0147] Step S360: Verify the network structure of the mutually fed-back corrected coupled-propagation causal network, check the continuity and rationality of the propagation path in the mutually fed-back corrected coupled-propagation causal network, and generate a coupled-propagation causal network with verification labels.

[0148] Specifically, various verification methods can be employed, such as simulation verification, which uses a physical model to simulate the coupled causal network after feedback correction and checks whether the simulation results meet expectations; and logical verification, which checks whether the logical relationships of the transmission paths are reasonable and whether there are any contradictions or irrationalities. If the continuity and rationality of the transmission paths are verified, a verification pass marker is added to the network; if problems exist, the network structure is modified until verification is passed. Through such verification operations, coupled causal networks with verification markers are generated, ensuring the accuracy and reliability of the network structure.

[0149] Step S370: Standardize the annotation of network nodes in the coupled propagation causal network with verification labels to ensure that the annotation format of each propagation path node is consistent, and generate the annotated coupled propagation causal network.

[0150] To facilitate the management and analysis of coupled causal networks with verification labels, the nodes in the network are standardized in their labeling. Each path node may have had different labeling formats in previous processing, which can cause inconvenience in subsequent operations. A unified labeling format rule is established, such as "[Node Type]-[Node Number]", where "[Node Type]" indicates the node type, such as chip node, substrate node, package node, etc., and "[Node Number]" is the node's unique identifier within the entire network. Each path node in the coupled causal network with verification labels is labeled according to this unified format, ensuring consistency across all nodes. This standardized labeling process generates the labeled coupled causal network.

[0151] Step S400: Drive the thermal coupling evolution model through the mutually fed-back corrected coupled conduction causal network, perform dynamic simulation of multi-stage photoelectric thermal conduction, and generate a coupled-driven thermal operation evolution probability map and a dynamic instability critical domain that is updated in real time as the simulation progresses.

[0152] The thermal coupling evolution model is used to simulate the interaction and evolution of light, electricity, and heat during the operation of a MiniLED display. It consists of three modules: physical parameters, photoelectric conversion, and heat conduction. The physical parameter module stores various physical parameters of the MiniLED display, including the thermal conductivity and specific heat capacity of the chip material, the thermal conductivity characteristics of the substrate, and the thermal resistance of the packaging material. These parameters are precisely set according to the actual design and manufacturing process of the display. The photoelectric conversion module simulates the process of the chip converting electrical energy into light energy, considering the relationship between the chip's electro-optical conversion efficiency and temperature and current. For example, as the chip junction temperature increases, the electro-optical conversion efficiency decreases, and the module performs calculations and simulations based on this rule. The heat conduction module is the core, simulating the conduction of heat between the chip, substrate, and packaging based on the laws of thermal conduction. It considers the differences in thermal conduction of different materials, contact thermal resistance, and other factors to construct a detailed heat conduction path model to accurately predict changes in chip junction temperature and heat distribution.

[0153] During the thermal coupling evolution model operation, the coupled conduction causal network after mutual feedback correction is used as input. In the early stage, based on initial conditions such as the initial chip current and ambient temperature, the photoelectric conversion module simulates electro-optical conversion and calculates the heat generated by the chip; the heat conduction module calculates the heat conduction inside the chip based on the heat generation and heat conduction characteristics, predicting the initial rise trend of the junction temperature. In the middle and late stages, as heat accumulation and thermal effects intensify, the calculation results of each module are updated in real time. When the chip junction temperature rises, the photoelectric conversion module adjusts the electro-optical conversion efficiency, and the heat conduction module considers the influence of junction temperature on the heat transfer coefficient to accurately simulate heat conduction and diffusion.

[0154] The model continuously iterates and updates the results of each module, realizing multi-stage dynamic simulation of photoelectric and thermal conduction, and generating real-time information on changes in physical quantities such as chip junction temperature, electro-optical conversion efficiency, and heat flow distribution.

[0155] In one implementation, step S400 may specifically include the following steps S410 to S460:

[0156] Step S410: Input the mutually fed-back modified coupled conduction causal network into the thermally coupled evolution model, so that the deduction logic of the thermally coupled evolution model and the conduction path of the mutually fed-back modified coupled conduction causal network are deeply related, ensuring that the deduction process follows the logic of the mutually fed-back modified coupled conduction causal network, and generating a set of bound deduction logic.

[0157] When inputting the mutually fed-back modified coupled conduction causal network into the thermally coupled evolution model, the deduction logic of the thermally coupled evolution model needs to be deeply correlated with the network's conduction paths. The mutually fed-back modified coupled conduction causal network details the conduction paths and relationships of light, electricity, and heat between various parts such as chips, substrates, and packages, and the deduction logic of the thermally coupled evolution model needs to be compatible with it.

[0158] In one implementation, step S410 includes the following steps S411 to S418:

[0159] Step S411: Extract all transmission paths and their relationships in the coupled causal network after mutual feedback correction, record the association methods of the starting node, ending node and intermediate nodes of each transmission path, and generate a transmission path association set.

[0160] The mutually fed-back corrected coupled conduction causal network contains the optical, electrical, and thermal conduction paths and relationships of MiniLED displays under the conditions of chip manufacturing discreteness and environmental thermal disturbances. In order to effectively correlate this network with the thermally coupled evolution model, it is necessary to extract all conduction paths and their relationships.

[0161] For each conduction path in the network, the relationships between its starting node, ending node, and intermediate nodes are recorded in detail. The starting node is the beginning of the conduction path, such as a heat source node on a chip; the ending node is the end of the conduction path, such as a heat dissipation area node on a substrate; intermediate nodes are other nodes in the conduction path that connect the starting and ending nodes, such as circuit nodes inside the chip or heat conduction channel nodes on the substrate. The relationships between intermediate nodes describe the interactions between nodes, such as the direction of heat conduction and the intensity of current conduction.

[0162] Taking heat conduction as an example, for a conduction path from the chip to the substrate, the chip node is recorded as the starting node and the substrate node as the ending node, with intermediate nodes possibly passing through the chip's heat dissipation pins and the heat conduction layer on the substrate. The direction of heat conduction is recorded as from the chip to the substrate, and the intensity of heat conduction is determined based on the conduction parameters in the coupled conduction causal network after mutual feedback correction. By recording and analyzing all conduction paths in the network in this way, a set of conduction path associations is generated. This set provides a detailed information foundation for subsequently matching the conduction paths with the deductive logic of the thermal coupling evolution model.

[0163] Step S412: Input the set of transmission path associations into the thermally coupled evolution model, perform association matching with the inference logic inside the thermally coupled evolution model, match the corresponding inference logic rules for each transmission path, and generate the matching result of path and logic.

[0164] The thermally coupled evolution model can be pre-defined with deductive logic rules to simulate the interaction and evolution of light, electricity, and heat, based on actual needs. These rules are grounded in physical laws and mathematical models, such as the laws of heat conduction and circuit laws. After inputting the set of conduction path associations into the thermally coupled evolution model, the conduction paths are matched with these deductive logic rules.

[0165] For each conduction path in the set of associated conduction paths, a corresponding deduction logic rule is matched based on its physical characteristics and relationships. For example, for a conduction path describing the conduction of current in the internal circuitry of a chip, a current conduction deduction logic rule based on circuit laws is matched. This rule calculates the magnitude and direction of the current based on parameters such as the circuit's resistance and voltage. For a conduction path describing the transfer of heat between the chip and the package, a heat conduction deduction logic rule is matched. This rule calculates the heat transfer process based on the laws of heat conduction and parameters such as the thermal conductivity coefficient in the conduction path.

[0166] During the matching process, factors such as the association methods of the starting node, ending node, and intermediate nodes of the conduction path are considered. For example, if the starting node of the conduction path is a high-voltage node and the ending node is a low-voltage node, then the matching current conduction deduction logic rules need to be able to correctly handle the current conduction process caused by this voltage difference. Through such matching operations, suitable deduction logic rules are found for each conduction path, generating the matching results between the path and the logic.

[0167] Step S413: Based on the matching results of paths and logic, bind each transmission path in the transmission path association set with the corresponding deduction logic rule to ensure that the deduction process of each transmission path follows the corresponding deduction logic rule, and generate a set of deduction logic after binding in which the transmission path and deduction logic are deeply associated.

[0168] After obtaining the matching results between the paths and the logic, each transmission path in the transmission path association set is bound to its corresponding deduction logic rule. The binding operation tightly links the transmission path and the deduction logic rule, ensuring that the deduction process of each transmission path strictly follows the corresponding deduction logic rule during the dynamic deduction of the thermally coupled evolution model.

[0169] For example, for a heat conduction path from the chip to the substrate, a heat conduction deduction logic rule has already been matched, and the conduction path is bound to this deduction logic rule. In subsequent deduction processes, when simulating the heat conduction process of this path, calculations are performed strictly according to the heat conduction deduction logic rule, and the heat transfer process between the chip and the substrate is accurately calculated based on the laws of heat conduction and the conduction parameters in the network.

[0170] Step S414: Perform a logical consistency check on the bound deduction logic set, eliminate logical conflicts and binding relationships that do not match the path and logic, and generate a logically consistent deduction logic set.

[0171] Within the bound set of deductive logic, there may be logical conflicts and path-logic mismatches. A logical conflict occurs when the bound deductive logic rules contradict the physical characteristics and relationships of the conduction path. For example, a conduction path represents heat transfer from a high temperature to a low temperature, but the matching deductive logic rules calculate heat transfer from a low temperature to a high temperature; this is a logical conflict. A path-logic mismatch occurs when the bound deductive logic rules cannot accurately describe the conduction process of the conduction path, such as using deductive logic rules for current conduction to handle heat conduction paths.

[0172] A logical consistency check is performed on the bound inference logic set using multiple methods. Logical analysis can be used to check whether the logical structure of the inference logic rules matches the physical logic of the transmission path; simulation verification can also be used, employing simple simulation scenarios to check whether the binding relationship produces reasonable results during the simulation process.

[0173] If logical conflicts or mismatches between paths and logic are found, these bindings are removed from the bound set of deductive logic. For example, if a propagation path and its bound deductive logic rule are found to have calculation results that do not conform to physical laws during simulation verification, the binding is removed. Through such checks and removals, a logically consistent set of deductive logic is generated.

[0174] Step S415: Adjust the logical order of the logically consistent deduction logic set to ensure that the order of the deduction logic is consistent with the temporal order of the transmission path, and generate the deduction logic set after the order adjustment.

[0175] In a logically consistent set of deduction logic, although each conduction path is correctly bound to its corresponding deduction logic rule, the order of the deduction logic may not be consistent with the temporal order of the conduction paths. The temporal order of the conduction paths reflects the order in which light, electricity, and heat are conducted in the actual process. For example, in a heat conduction process involving the chip to the substrate and then to the package, heat is first transferred from the chip to the substrate, and then from the substrate to the package. This is the temporal order of the conduction paths.

[0176] The logical order of the consistent deduction logic set is adjusted to align with the temporal order of the conduction paths. First, the temporal relationships of the conduction paths in the conduction path association set are analyzed to determine the order of each path. Then, based on this order, the execution order of the corresponding deduction logic rules is adjusted. For example, for a conduction path describing heat conduction from the chip to the substrate and a conduction path describing heat conduction from the substrate to the package, the deduction logic rule bound to the chip-to-substrate heat conduction path is executed before the deduction logic rule bound to the substrate-to-package heat conduction path. This logical order adjustment operation generates a deduction logic set with the adjusted order.

[0177] Step S416: Verify the binding results of the deduction logic set after the order adjustment, check the rationality and accuracy of the binding relationship, make the deduction logic and the transmission path completely match, and generate the bound deduction logic set.

[0178] Verifying the binding results of the rearranged deduction logic set is to ensure the rationality and accuracy of the binding relationship, and to ensure that the deduction logic and the transmission path are fully matched. Multiple verification methods are employed, such as simulation verification and comparative verification.

[0179] Simulation verification utilizes a thermally coupled evolution model to perform simple simulation scenarios and check whether the bonding relationships produce reasonable results during the simulation. For example, given a heat transfer path describing heat conduction from the chip to the substrate and its bonding heat conduction deduction logic rules, parameters such as the initial temperature of the chip and the initial temperature of the substrate are input into the simulation scenario. The simulation is then checked to see if the heat transfer process obtained conforms to physical laws and the conduction parameters in the network. If the simulation results are unreasonable, it indicates that there may be a problem with the bonding relationships, which need to be corrected.

[0180] Comparative verification involves comparing the bound set of deduction logic with the set of conduction path associations to check whether the deduction logic rules accurately reflect the physical characteristics and relationships of the conduction paths. For example, it checks whether parameters such as thermal conductivity and current conduction intensity used in the deduction logic rules are consistent with the parameters recorded in the set of conduction path associations.

[0181] If any unreasonable or inaccurate binding relationships are found, they are corrected until verification is successful. Through such verification and correction operations, the deduction logic is made to perfectly match the transmission path, generating a bound set of deduction logic.

[0182] Step S417: Add binding identifiers to the bound deduction logic set, add a unique binding identifier to each binding relationship, and generate a deduction logic set with identifiers.

[0183] To facilitate the management and tracking of binding relationships within the bound inference logic set, binding identifiers are added. A unique binding identifier is added to each binding relationship, for example, using the format "BL-[binding number]", where "BL" represents the binding relationship and "[binding number]" is a unique identifier for that binding relationship within the entire set.

[0184] By adding binding identifiers, each binding relationship has a clear identifier, facilitating querying, analysis, and management in subsequent operations. For example, during dynamic simulation, if a problem is found in a binding relationship, the binding identifier can be used to quickly locate the relationship for further inspection and correction. This identifier-adding operation generates a set of identifiable simulation logic, improving the manageability and efficiency of the entire thermal runaway early warning system.

[0185] Step S418: Verify the logical identifiers of the labeled deduction logic set to ensure that each binding identifier is completely consistent with the corresponding propagation path and deduction logic rules, and generate the verified deduction logic set.

[0186] Verifying the logical identifiers of the labeled inference logic set ensures that each binding identifier accurately corresponds to the propagation path and inference logic rules. The consistency between the binding identifiers, propagation paths, and inference logic rules is checked using various verification methods, such as data comparison and simulation verification.

[0187] Data comparison involves comparing the binding identifier with relevant information in the set of associated conduction paths and the set of deduction logic rules. This checks whether the binding identifier matches the association methods of the start, end, and intermediate nodes of the conduction path, as well as the specific content of the deduction logic rules. For example, it checks whether the heat conduction direction of the conduction path corresponding to the binding identifier is consistent with the heat conduction direction calculated in the deduction logic rules.

[0188] Simulation verification utilizes a thermal coupling evolution model to simulate the process and check whether the binding identifiers match the actual conduction paths and the execution of the deduced logic rules. For example, in simulating the heat conduction from the chip to the substrate, it checks whether the binding relationships corresponding to the binding identifiers correctly execute the heat conduction deduced logic rules.

[0189] If a binding identifier is found to be inconsistent with its corresponding propagation path and deduction logic rules, the binding identifier is corrected until verification passes. Through this verification and correction process, each binding identifier is made completely consistent with its corresponding propagation path and deduction logic rules, generating a verified deduction logic set.

[0190] Step S420: Based on the bound deduction logic set, perform dynamic deduction of multi-stage photoelectric thermal conduction, simulate the complete conduction process of photoelectric thermal effect in each thermal runaway process stage, record the deduction details of each thermal runaway process stage, and generate a multi-stage deduction result sequence.

[0191] After obtaining the validated set of deduction logic, a multi-stage dynamic deduction of photoelectric thermal conduction is performed based on this set. The thermally coupled evolution model simulates the interaction and conduction processes of light, electricity, and heat within each stage of thermal runaway, according to the bound set of deduction logic. The physical properties and interaction modes of light, electricity, and heat change at different stages of thermal runaway. For example, in the early stage, the chip generates relatively little heat, the electro-optical conversion efficiency is high, and heat conduction is mainly concentrated inside the chip; in the middle stage, the chip junction temperature rises, the electro-optical conversion efficiency decreases, and heat conduction begins to diffuse towards the substrate and package; in the late stage, the thermal runaway phenomenon intensifies, and the interaction of light, electricity, and heat becomes more complex.

[0192] The thermally coupled evolution model performs detailed simulations of each stage of the thermal runaway process, following the bound set of deductive logic. For each stage, it records the complete conduction process of photoelectric and thermal interactions, including heat generation, transfer, and dissipation; current conduction and changes; and light emission and absorption. For example, in simulating the heat conduction from the chip to the substrate, it records information such as the temperature changes of the chip and substrate and the rate of heat transfer at each time point.

[0193] Record the simulation details of each stage of thermal runaway, including the numerical changes, trends, and interactions of physical quantities, and generate a multi-stage simulation result sequence.

[0194] Step S430: Integrate the multi-stage simulation results sequence into a graph, arrange the simulation results of each thermal runaway process stage in chronological order, and generate a coupled-driven thermal operation evolution probability graph.

[0195] The multi-stage simulation results sequence contains detailed simulation results for each stage of the thermal runaway process. To present these results more intuitively, they are graphically integrated. The simulation results for each stage of the thermal runaway process are arranged in chronological order to construct a probability map of coupled-driven thermal evolution. In the map, the horizontal axis represents time, and the vertical axis represents the values ​​or probabilities of physical quantities such as light, electricity, and heat. For each stage of the thermal runaway process, the numerical values ​​and probability information of the physical quantities are mapped onto the map. For example, for the physical quantity of chip junction temperature, the map plots the numerical change curves of the chip junction temperature at different time points, and colors or shading can be used to represent the probability of the junction temperature being in different ranges. The electro-optical conversion efficiency is plotted in a similar way, showing its variation curve and probability distribution.

[0196] Step S440: Identify unstable regions in the coupled-driven thermal operation evolution probability map, track potential unstable regions that appear in the simulation process in real time, record the location and evolution trend of unstable regions, and generate dynamic unstable critical domains that are updated in real time with the simulation process.

[0197] The coupled-driven thermal runaway evolution probability map illustrates the dynamic changes and probability distributions of light, electricity, and heat in a MiniLED display during different stages of thermal runaway. Identifying instability regions in this map is crucial for timely detection of areas where the system may enter a thermal runaway state.

[0198] An unstable region is a region in the spectrum where changes in physical quantities such as light, electricity, and heat exceed normal ranges, potentially leading to system instability. For example, when the chip junction temperature exceeds a certain threshold, or the electro-optical conversion efficiency drops to an excessively low level, the system may enter an unstable region. Real-time tracking of potential unstable regions during the simulation process is achieved by setting a series of judgment rules and thresholds. For instance, if the chip junction temperature in the spectrum exceeds T1, the region is marked as a potential unstable region. Simultaneously, the location of the unstable region is recorded, i.e., its time and physical quantity coordinates in the spectrum; the evolution trend of the unstable region is also recorded, such as whether it is expanding or shrinking, and whether the degree of instability is aggravating or mitigating. As the simulation progresses, the identification results of unstable regions are continuously updated, generating a dynamic unstable critical domain that updates in real time with the simulation process.

[0199] Step S450: Link and store the coupled-driven thermal operation evolution probability map and dynamic instability critical domain, add corresponding inference time series markers to each inference result, and generate a set of inference results with inference time series markers.

[0200] For example, for each data point in the coupled-driven thermal runaway evolution probability map and each instability region in the dynamic instability critical domain, a corresponding projection time series marker is added. This marker clarifies the position of each projection result on the time axis, facilitating subsequent time series analysis and comparison. The coupled-driven thermal runaway evolution probability map and dynamic instability critical domain, with added projection time series markers, are then stored together to generate a set of projection results with projection time series markers. The data in this set is ordered and clearly correlated, improving the manageability and efficiency of the entire thermal runaway early warning system.

[0201] Step S460: Verify the simulation results set with simulation time sequence markers, check the logical consistency and accuracy of the coupled-driven thermal operation evolution probability map and dynamic instability critical domain, ensure the reliability of the simulation results, generate a simulation result set with verification markers, unify the storage markers of the simulation result set with verification markers, so that the storage marker format of each simulation result is consistent, and generate a simulation result set with unified markers.

[0202] The verification of the deduction results is to ensure the logical consistency and accuracy of the coupled-driven thermal operation evolution probability map and the dynamic instability critical region. Multiple verification methods are employed, such as physical verification and comparative verification. Physical verification checks whether the deduction results conform to physical laws and actual conditions. For example, it checks whether the changes in physical quantities such as light, electricity, and heat in the coupled-driven thermal operation evolution probability map conform to physical laws such as the heat conduction law and circuit laws. Comparative verification compares the deduction results with actual experimental data or historical data to check the accuracy of the deduction results. For example, it compares the chip junction temperature change curve in the map with the actual measured chip junction temperature data to check for significant deviations. If logical inconsistencies or inaccuracies are found in the deduction results, they are corrected until verification is passed. Through such verification and correction operations, a verification-passed identifier is added to the deduction results, generating a set of deduction results with verification identifiers. To facilitate the management and querying of the set of deduction results with verification identifiers, their storage identifiers are standardized. Establish a unified storage identifier format rule, such as adopting the format "RR-[Storage Number]", where "RR" represents the deduction result and "[Storage Number]" is the unique number of the deduction result in the entire set.

[0203] Step S500: Based on the coupled-driven thermal operation evolution probability map and dynamic instability critical domain, combined with the real-time sampling records of the MiniLED display screen, conduction causal chain matching and early warning judgment are performed to complete the thermal runaway early warning of the MiniLED display screen.

[0204] In one implementation, step S500 may specifically include the following steps S510 to S570:

[0205] Step S510: Time-align the coupled-driven thermal operation evolution probability map and dynamic instability critical region with the real-time sampling record of the MiniLED display screen, so that the timing nodes of the real-time sampling record of the MiniLED display screen are completely consistent with the timing nodes of the coupled-driven thermal operation evolution probability map, ensuring the accuracy of the matching timing, and generating the timing alignment result.

[0206] To accurately match the real-time sampling records of the MiniLED display with the coupled-driven thermal evolution probability map and dynamic instability critical region, a timing alignment operation is required. First, all timing node information in the coupled-driven thermal evolution probability map is extracted, including the timestamp of each timing node, the corresponding thermal runaway process stage, and the map region range, generating a map timing node set containing complete node attributes. Then, all timing node information in the real-time sampling records of the MiniLED display is extracted, including the timestamp of each real-time sampling record, the corresponding chip-substrate-package coupling node, and the sampling content, generating a real-time sampling timing node set containing complete sampling attributes. The map timing node set and the real-time sampling timing node set are time-matched to determine the timing nodes with the same or closest times in both sets, and the matching correspondence is recorded, generating the node matching result. For example, if a timing node in the map has a time of t1, and a node in the real-time sampling record with a time closest to t1 has a time of t2, these two nodes are matched, and the correspondence is recorded.

[0207] Based on the node matching results, each real-time sampling record of the MiniLED display screen in the real-time sampling record is associated with the spectral region of the corresponding time-series node in the coupled-driven thermal operation evolution probability map, and the real-time sampling record is mapped to the corresponding spectral position to generate the association result between sampling and spectral.

[0208] The correlation results between sampling and the graph are adjusted for time sequence alignment. Independent time interpolation is performed on the time series of different physical quantities in the real-time sampling records of the MiniLED display. Time interpolation ensures that the timestamps of each physical quantity data point after interpolation match the time sequence nodes of the coupled-driven thermal evolution probability graph. For example, for the real-time sampling record of the chip junction temperature, a linear interpolation method is used to insert appropriate data points between adjacent sampling data points, ensuring that the timestamps of the interpolated chip junction temperature data points match the time sequence nodes of the graph. This interpolation process ensures accurate timing matching and generates time sequence alignment results.

[0209] The timing alignment results are verified for accuracy by checking the temporal consistency between the real-time sampling records of the MiniLED display and the spectral regions of the coupled-drive thermal evolution probability map. Verification can be performed through comparison to check whether the physical quantity values ​​of the real-time sampling records and the spectral regions reasonably match at the same time point; alternatively, simulation verification can be used to check whether the matching results after timing alignment meet expectations. If inaccurate alignment is found, the timing alignment results are corrected until verification is successful. This verification process generates timing alignment results with verification indicators.

[0210] To facilitate the management and tracking of timing alignment results, alignment identifiers are added to timing alignment results with verification flags, giving each alignment result a unique alignment identifier. For example, the format "TA-[alignment number]" can be used as the alignment identifier, where "TA" represents timing alignment and "[alignment number]" is the unique number of the alignment result in the entire set.

[0211] The identified time-series alignment results are stored, and a storage location identifier is added to each identified result. This identifier can be a storage address in the database or a file path, facilitating subsequent querying and use of the time-series alignment results. This storage operation generates a set of stored time-series alignment results.

[0212] Step S520: Based on the timing alignment results, associate and match the real-time sampling records of the MiniLED display screen with the corresponding spectral regions in the thermal operation evolution probability map of the coupled drive, map each real-time sampling record of the MiniLED display screen to the corresponding spectral region, and generate a set of matching results between the sampling records and the spectral regions.

[0213] After obtaining the timing alignment result, the real-time sampling records of the MiniLED display are associated and matched with the corresponding spectral regions in the coupled-driven thermal operation evolution probability map. Since the timing is already aligned, each real-time sampling record can find a corresponding spectral region. For each real-time sampling record of the MiniLED display, the spectral region of the corresponding timing node in the coupled-driven thermal operation evolution probability map is found according to its time stamp and physical quantity information. For example, for a real-time sampling record of the chip junction temperature collected at time t1, the spectral region corresponding to time t1 is found in the map, and the sampling record is mapped to this spectral region. By performing such an association and matching operation on all real-time sampling records of the MiniLED display, each sampling record is mapped to the corresponding spectral region, generating a set of matching results between sampling records and spectral regions.

[0214] Step S530: Perform correlation identification of the causal chain of the matching result set of the sampling record and the map region, extract the causal chain that matches the real-time sampling record of the MiniLED display, record the transmission path and correlation relationship of all causal chains, and generate a set of causal chains for correlation identification.

[0215] The matching set of sampling records and spectral regions determines the corresponding position of each MiniLED display's real-time sampling record in the coupled-driven thermal operation evolution probability map. This set is then used for conduction causal chain identification, extracting conduction causal chains that match the real-time sampling records from the spectral regions. Each spectral region contains a series of conduction causal chains describing the conduction processes and interrelationships of light, electricity, and heat within that region. Based on the physical quantities of the real-time sampling records, such as chip junction temperature and electro-optical conversion efficiency, matching conduction causal chains are selected from the spectral regions. For example, if the real-time sampling records show an increase in chip junction temperature, conduction causal chains describing the increase in chip junction temperature are extracted from the spectral regions. The conduction paths and relationships of all extracted conduction causal chains are recorded, including the start node, end node, and the association methods of intermediate nodes. Through this recording operation, a set of identified causal chains is generated.

[0216] Step S540: Compare the set of causal chains identified by association with the dynamic instability critical region, check whether the causal chains identified by association have entered the dynamic instability critical region, record the detailed comparison results, and generate a set of comparison results.

[0217] The set of causal chains identified through correlation analysis includes conducted causal chains that match the real-time sampling records of the MiniLED display. The dynamic instability critical region is the area where the system may enter a state of thermal runaway. The set of causal chains identified through correlation analysis is compared with the dynamic instability critical region to check whether these conducted causal chains have entered the dynamic instability critical region.

[0218] For each causal chain in the identified set of causal chains, its transmission path and physical quantity information are used to determine whether it intersects with the dynamic instability critical region. For example, if the value of a physical quantity in a causal chain exceeds the range of that physical quantity in the dynamic instability critical region, it indicates that the causal chain has entered the dynamic instability critical region. Detailed comparison results are recorded, including the causal chain number, the judgment result of entering or not entering the dynamic instability critical region, and the time point of entry. This recording process generates a comparison result set.

[0219] Step S550: Based on the comparison result set, perform a thermal runaway early warning judgment for the MiniLED display screen. If the causal chain of the associated identification enters the dynamic instability critical domain, an early warning signal is generated; otherwise, no early warning signal is generated, and a clear early warning judgment result is generated.

[0220] The comparison result set records the relationship between the identified causal chains and the dynamic instability critical region. Based on this set, a thermal runaway early warning judgment is made for the MiniLED display. If any identified causal chain in the comparison result set enters the dynamic instability critical region, it indicates that the MiniLED display may be about to enter a thermal runaway state, generating an early warning signal; if none of the identified causal chains enter the dynamic instability critical region, no early warning signal is generated. Through this early warning judgment operation, a clear early warning judgment result is generated. This result provides an important decision-making basis for taking timely measures to prevent thermal runaway of the MiniLED display. If an early warning signal is generated, methods such as reducing display brightness and increasing heat dissipation measures can be taken to avoid thermal runaway; if no early warning signal is generated, the system's operating status can continue to be monitored.

[0221] Step S560: Convert the early warning judgment result into a standardized early warning instruction, generate an executable early warning instruction according to the preset format, complete the thermal runaway early warning for the MiniLED display screen, verify the instruction format of the standardized early warning instruction to ensure that the standardized early warning instruction format meets the preset requirements, and generate the verified standardized early warning instruction.

[0222] The preset format may require the warning instruction to include information such as the warning level, the reason for the warning, and the recommended measures. If the warning judgment result is to generate a warning signal, then when generating a standardized warning instruction, the warning level will be determined according to the specific situation, such as a mild warning, a moderate warning, or a severe warning; the cause of the warning will be analyzed, such as excessively high chip junction temperature or low electro-optical conversion efficiency; and corresponding recommended measures will be given, such as reducing display brightness or increasing cooling fan speed. This information will be organized according to the preset format to generate an executable warning instruction.

[0223] After generating standardized early warning commands, their format is verified. The verification process involves a detailed comparison between the generated command and the preset format, checking whether the command contains all necessary information, whether the information format is correct, and whether the content of each field meets requirements. For example, it checks whether the warning level is within the specified range and whether the recommended measures are feasible. If the command format is found to be inconsistent with the preset requirements, it is corrected until verification is successful. Through this verification process, standardized early warning commands are generated, ensuring that the commands can be accurately recognized and executed by relevant systems or equipment, thereby effectively completing the thermal runaway early warning task for MiniLED displays.

[0224] Step S570: Associate the verified standardized warning instructions with the corresponding real-time sampling records of the MiniLED display screen, add the corresponding real-time sampling record identifier of the MiniLED display screen to each standardized warning instruction, and generate a set of associated warning instructions.

[0225] The validated standardized early warning instructions contain important information about thermal runaway warnings for MiniLED displays. To facilitate subsequent querying, analysis, and tracing, these instructions need to be associated and stored with the corresponding real-time sampling records of the MiniLED displays.

[0226] Each standardized warning command is generated based on corresponding real-time sampling records, which reflect the actual physical quantities of the MiniLED display screen, such as light, electricity, and heat, at a specific moment. A corresponding real-time sampling record identifier for the MiniLED display screen is added to each standardized warning command. This identifier can be unique information such as the timestamp or sampling number of the real-time sampling record. By adding this identifier, a correlation is established between the warning command and the real-time sampling record.

[0227] For example, if a standardized warning command is generated due to excessively high chip junction temperature at a certain moment, then a real-time sampling record identifier for that moment is added to the warning command. All standardized warning commands with added identifiers are stored, generating an associated set of warning commands. Each warning command in this set is associated with a corresponding real-time sampling record, facilitating quick retrieval of the corresponding real-time sampling data when needed later. This allows for further analysis of the causes and trends of thermal runaway, providing strong data support for optimizing thermal management strategies and improving display design. Simultaneously, the associated storage method improves the manageability and data utilization efficiency of the entire thermal runaway warning system, enabling the system to more efficiently address thermal runaway issues in MiniLED displays.

[0228] This application generates a multi-scale optoelectronic-thermal time-series dataset anchored by synchronously collecting three types of parameters: chip junction temperature fluctuation, electro-optical conversion efficiency decay, and substrate thermal conduction differences, and binding them to conduction link markers of chip-substrate-package coupling nodes. Based on this dataset, a causal convolutional network is used to mine the conduction causal chains between each stage of thermal runaway. By reverse verification, pseudo-correlated sampling records are removed to accurately capture the real conduction logic. Subsequently, using the purified causal chain as the core, bidirectional mutual feedback calibration is performed in combination with chip manufacturing discreteness and environmental thermal disturbance sampling records to remove conduction deviation causal chains and construct a coupled conduction causal network that fits the actual operating scenario. This network then drives a thermal coupling evolution model to generate a thermal operation evolution probability map and dynamic instability critical domain, fully presenting the gradual evolution process of thermal runaway. Finally, based on the real-time sampling of conduction causal chains matched with the map and critical domain, an early warning is completed. This identifies thermal runaway risks from the conduction logic level, improves the accuracy and foresight of the early warning, and effectively enhances the ability to prevent and control MiniLED thermal runaway risks.

[0229] Based on the foregoing embodiments, this invention provides an early warning device. The units and modules included in the device can be implemented by a processor in a computer device; of course, they can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0230] Figure 2 This is a schematic diagram of the composition structure of an early warning device provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the early warning device 200 includes:

[0231] The data acquisition module 210 is used to synchronously start the MiniLED display multi-scale sensing node to collect continuous sampling records of chip junction temperature fluctuation, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operation cycle. Through the coupling conduction link marking logic built into the sensing node, the corresponding chip-substrate-package coupling node conduction link mark is bound to each continuous sampling record to obtain a multi-scale optoelectronic thermal time series dataset anchored to the conduction link.

[0232] The causal mining module 220 is used to mine the conduction causal chain of photoelectric and thermal effects between different stages of thermal runaway based on the multi-scale photoelectric and thermal time series dataset based on anchored conduction links. It uses a causal convolutional network to mine the conduction causal chain of photoelectric and thermal effects between different stages of thermal runaway process. It also uses reverse causal verification logic to remove pseudo-associative sampling records and obtain a set of cross-stage coupled dominant causal chains.

[0233] The mutual feedback calibration module 230 is used to perform bidirectional mutual feedback calibration by taking the set of cross-stage coupling dominant causal chains as the core input, introducing continuous sampling records of chip manufacturing discreteness and environmental thermal disturbances, and removing the conduction deviation causal chains generated during the calibration process through the causal chain deviation conduction identification logic to obtain the coupled conduction causal network after mutual feedback correction.

[0234] The dynamic simulation module 240 is used to drive the thermal coupling evolution model through the mutually fed-back corrected coupled conduction causal network, perform dynamic simulation of multi-stage photoelectric thermal conduction, generate a coupled-driven thermal operation evolution probability map and a dynamic instability critical domain that is updated in real time with the simulation process.

[0235] The runaway early warning module 250 is used to perform conduction causal chain matching and early warning judgment based on the thermal operation evolution probability map and dynamic instability critical domain driven by coupling, combined with the real-time sampling record of the MiniLED display, to complete the thermal runaway early warning of the MiniLED display.

[0236] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. In some embodiments, the functions or modules included in the device provided by the present invention can be used to execute the methods described in the above method embodiments. For technical details not disclosed in the device embodiments of the present invention, please refer to the descriptions of the method embodiments of the present invention for understanding. It should be noted that in the embodiments of the present invention, if the above-mentioned MiniLED display thermal runaway early warning method based on the coupling model is implemented in the form of software functional modules and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the related technology, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of the present invention are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.

Claims

1. A method for thermal runaway early warning of a Mini LED display screen based on a coupling model, characterized in that, The method includes: Simultaneously start the MiniLED display multi-scale sensing node to collect continuous sampling records of chip junction temperature fluctuation, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operation cycle. Through the coupling conduction link marking logic built into the sensing node, bind the corresponding chip-substrate-package coupling node conduction link mark to each continuous sampling record to obtain a multi-scale optoelectronic thermal time series dataset anchored to the conduction link. Based on the multi-scale photoelectric and thermal time-series dataset of the anchored conduction link, the conduction causal chain of photoelectric and thermal interaction between different thermal runaway process stages is mined based on the causal convolutional network. The pseudo-associative sampling records are stripped off by the reverse causal verification logic to obtain the set of cross-stage coupled dominant causal chains. Using the cross-stage coupling dominant causal chain set as the core input, continuous sampling records of chip manufacturing discreteness and environmental thermal disturbance are introduced for bidirectional mutual feedback calibration. The conduction deviation causal chain generated during the calibration process is removed through the causal chain deviation conduction identification logic to obtain the coupled conduction causal network after mutual feedback correction. The coupled conduction causal network after mutual feedback correction drives the thermal coupling evolution model, performs dynamic simulation of multi-stage photoelectric thermal conduction, and generates a probability map of coupled thermal operation evolution and a dynamic instability critical domain that is updated in real time with the simulation process. Based on the coupled-driven thermal operation evolution probability map and dynamic instability critical domain, combined with the real-time sampling records of the MiniLED display screen, a conduction causal chain matching and early warning judgment is performed to complete the thermal runaway early warning of the MiniLED display screen.

2. The method of claim 1, wherein, The synchronous startup of the MiniLED display multi-scale sensing node collects continuous sampling records of chip junction temperature fluctuations, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operating cycle. Through the built-in coupling conduction link marking logic of the sensing node, each continuous sampling record is bound to a corresponding chip-substrate-package coupling node conduction link mark, resulting in a multi-scale optoelectronic-thermal time-series dataset anchored to the conduction link, including: The MiniLED display multi-scale sensing nodes are synchronously started to collect continuous sampling records of chip junction temperature fluctuation, electro-optical conversion efficiency decay, and substrate thermal conduction difference according to the time sequence nodes of the entire operation cycle. Each continuous sampling record is bound one-to-one with the operation status record of the corresponding time sequence node to generate a sampling record sequence containing time sequence markers and operation status association information. The conduction link is traced for the time-bound sampling record sequence to track the complete conduction path of the chip-substrate-package coupling node corresponding to each continuous sampling record, and a conduction path tracing result containing detailed information of the path start node and end node is generated; Based on the source tracing results of the conduction path, the conduction link marker of the corresponding conduction path is bound to each segment of the continuous sampling record, and the continuous sampling records under the same conduction path are concatenated in chronological order to generate a set of sampling records bound to the conduction path that is deeply associated with the continuous sampling records; The sampling record set bound to the transmission path is arranged in chronological order of the entire operation cycle. Based on the timestamp of the same time node, an association index is established for the continuous sampling records under different transmission paths to generate a cross-path time-related sampling record set. The cross-path time-series associated sampling record set is unified and integrated with the conduction link label, so that the conduction link label format of all the continuous sampling records is consistent, and the multi-scale photoelectric and thermal time-series dataset anchored to the conduction link is generated. The multi-scale photoelectric and thermal time-series dataset of the anchored conduction link is grouped and stored according to the conduction path dimension, generating an independent storage group corresponding to each conduction path; The independent storage groups are uniformly labeled with transmission path identifiers so that the identifier of each independent storage group is consistent with the format of the transmission link mark, thereby generating a labeled group storage set.

3. The method of claim 2, wherein, The step of arranging the sampling record set bound to the transmission path in chronological order of the entire operation cycle, establishing an association index for the continuous sampling records under different transmission paths based on the timestamp of the same time node, and generating a cross-path time-related sampling record set includes: Extract the time sequence markers of each continuous sampling record from the sampling record set bound to the conduction path, sort all the time sequence markers in chronological order, and generate an ordered time sequence marker sequence containing all time nodes; Each continuous sampling record in the sampling record set bound to the conduction path is matched one by one with the corresponding time marker in the ordered time marker sequence to generate a time-ordered sampling record set. Extract all consecutive sampling records under the same time node from the time-ordered sampling record set, and generate multi-path sampling record groups for the same time node based on their common timestamp; Add a unified timestamp identifier based on the time sequence node to the multi-path sampling record group of the same time sequence node to generate a sampling record group with timestamp identifier; All sampling records with cross-path identifiers are grouped and arranged in chronological order to generate a set of sampling records with cross-path chronological association; Perform a continuity check on the cross-path temporal associated sampling record set, remove the sampling record groups with cross-path identifiers that have temporal breaks, and generate a continuous and complete cross-path temporal associated sampling record set. The continuous and complete cross-path temporal correlation sampling record set is standardized by unifying the cross-path correlation identifiers, so that the format of all cross-path correlation identifiers is consistent, and a standardized cross-path sampling record set is generated.

4. The method of claim 1, wherein, The multi-scale photoelectric-thermal time-series dataset based on the anchored conduction links uses a causal convolutional network to mine the conduction causal chains of photoelectric-thermal interactions between different stages of thermal runaway. By using reverse causal verification logic to remove pseudo-associative sampling records, a set of cross-stage coupled dominant causal chains is obtained, including: The multi-scale photoelectric and thermal time-series dataset of the anchored conduction links is input into the causal convolutional network to mine the conduction causal chains of the photoelectric and thermal effects between different stages of the thermal runaway process, and generate an initial conduction causal chain sequence containing all potential related links. The initial conduction causal chain sequence is divided into the thermal runaway process stages, and each conduction causal chain is assigned to the corresponding thermal runaway process stage according to the time node range it covers, generating a multi-stage conduction causal chain sub-sequence; The causal direction of each stage-transmission causal chain subsequence is verified. The rationality of the causal direction of each transmission causal chain is checked through the reverse causal verification logic, and a causal direction verification result containing the verification result of each transmission causal chain is generated. Based on the causal direction verification results, the pseudo-correlation links with unreasonable causal directions in the stage-transmission causal chain sub-sequence are removed to generate the stage-dominant causal chain sequence. The multi-stage dominant causal chain subsequences are cross-stage correlated and integrated, and the conduction causal chains that have conduction correlation between adjacent thermal runaway process stages are connected in series to generate a cross-stage coupled dominant causal chain set. The set of dominant causal chains in cross-stage coupling is sorted by causal relationship from high to low, generating a set of causal chains sorted by the degree of causal relationship. The set of causal chains sorted by correlation strength is labeled with a unified correlation strength label, so that the correlation strength label format of each causal chain is consistent, and a labeled set of causal chains is generated.

5. The method of claim 4, wherein, The initial causal chain sequence is divided into stages of the thermal runaway process. Each causal chain is assigned to a corresponding thermal runaway process stage according to the time-series nodes it covers, generating multiple stage-based causal chain sub-sequences, including: Extract the start and end time nodes of each of the initial causal chain sequences, record the time range covered by each causal chain, and generate a causal chain time boundary set containing the time boundary information of each causal chain. Extract the time sequence node range of each stage of the thermal runaway process within the entire operating cycle, record the start and end time points of each thermal runaway process stage, and generate a stage time sequence range set containing the complete time range of each thermal runaway process stage. The temporal boundary of each of the conduction causal chains in the set of temporal boundaries of the causal chains is matched with the temporal range of the corresponding thermal runaway process stage in the set of temporal ranges of the stages to determine the thermal runaway process stage to which each conduction causal chain belongs, and a matching result between the conduction causal chain and the thermal runaway process stage is generated. Based on the matching results between the conduction causal chain and the thermal runaway process stage, the conduction causal chains belonging to the same thermal runaway process stage are grouped together to generate a stage causal chain grouping set corresponding to each thermal runaway process stage. The conduction causal chains in each stage causal chain grouping set are integrated sequentially, and the sequentially continuous conduction causal chains within the same thermal runaway process stage are connected in series to form a complete conduction link within the thermal runaway process stage, generating a continuous stage causal chain subsequence. Bind multiple consecutive stage causal chain sequences to the thermal runaway process stage identifier, add a unique identifier corresponding to the thermal runaway process stage to each consecutive stage causal chain sequence, and generate multiple stage conduction causal chain sequences. The stage identifiers of the multiple stage-based causal chain sequences are verified to ensure that the identifier of each stage-based causal chain sequence is completely consistent with the corresponding thermal runaway process stage, thereby generating a set of verified stage-based causal chain sequences.

6. The method of claim 5, wherein, Based on the matching results between the conduction causal chain and the thermal runaway process stage, the conduction causal chains belonging to the same thermal runaway process stage are grouped together to generate a stage causal chain grouping set corresponding to each thermal runaway process stage, including: Extract all the conduction causal chains corresponding to each thermal runaway process stage from the matching results of the conduction causal chain and the thermal runaway process stage, summarize the conduction path and temporal boundary information of all the conduction causal chains, and generate a candidate set of causal chains for each thermal runaway process stage; For each thermal runaway process stage, the similarity of the conduction path of the candidate causal chain is compared, and the conduction causal chains with similar conduction path structures and the same active nodes are clustered to generate a causal chain cluster subset within the thermal runaway process stage. The causal chain cluster subsets within each thermal runaway process stage are integrated by conducting the transmission associations. The transmission causal chains with direct transmission associations in the causal chain cluster subsets are connected in series to form complete transmission links in the causal chain cluster subsets, generating an integrated stage causal chain subset. All the integrated stage causal chain subsets within the same thermal runaway process stage are grouped and integrated, and all the integrated stage causal chain subsets within the thermal runaway process stage are merged into a complete set to generate the stage causal chain grouping set corresponding to the thermal runaway process stage. The grouping results of each stage causal chain grouping set are verified. It is checked whether the grouped causal chains all belong to the corresponding thermal runaway process stage. Erroneously grouped links that do not belong to the thermal runaway process stage are eliminated, and grouping verification results are generated. Based on the grouping verification results, the grouping set of the stage causal chains is adjusted to ensure that all the conduction causal chains in the set belong to the corresponding thermal runaway process stage, thus generating the final grouping set of stage causal chains. Add grouping identifiers to the final stage causal chain grouping set, and add a grouping identifier corresponding to the thermal runaway process stage to each stage causal chain grouping set to generate a stage causal chain grouping set with grouping identifiers.

7. The method according to claim 1, characterized in that, The process involves using the cross-stage coupled dominant causal chain set as the core input, incorporating continuous sampling records of chip manufacturing discreteness and environmental thermal disturbances for bidirectional mutual feedback calibration, and removing the conduction deviation causal chains generated during the calibration process through causal chain deviation propagation identification logic to obtain the mutually fed-back corrected coupled conduction causal network, including: Using the cross-stage coupling dominant causal chain set as the core input, continuous sampling records of chip manufacturing discreteness are introduced. Based on the manufacturing discrete information, the correlation relationship or state parameters of the relevant transmission path nodes in the transmission causal chain are corrected, and mutual feedback calibration in the first direction is performed to generate the first calibrated causal chain set. Using the first calibrated causal chain set as input, continuous sampling records of environmental thermal disturbances are introduced. Based on the thermal disturbance information, the correlation relationship or state parameters of the relevant conduction path nodes in the conduction causal chain are corrected again, and mutual feedback calibration in the second direction is performed to generate the second calibrated causal chain set. The second calibrated causal chain set is subjected to deviation propagation identification. The propagation deviation causal chain generated during the calibration process is detected by the causal chain deviation propagation identification logic, and a deviation propagation causal chain set containing all propagation deviation causal chains is generated. The second calibrated causal chain set is compared with the deviation transmission causal chain set. The transmission deviation causal chains that belong to the deviation transmission causal chain set in the second calibrated causal chain set are removed, and only the valid transmission causal chains that are calibrated reasonably are retained to generate a corrected causal chain set. The modified causal chain set is integrated across stages, and all the modified transmission causal chains are connected in series according to the natural association of the transmission path to form a complete transmission network covering the entire operation cycle, generating a mutually feed-back modified coupled transmission causal network. The network structure of the mutually fed-back corrected coupled causal network is verified, and the continuity and rationality of the transmission paths in the mutually fed-back corrected coupled causal network are checked to generate a coupled causal network with verification labels. The coupled causal network with verification identifiers is labeled in a unified manner to ensure that the labeling format of each transmission path node is consistent, thus generating the labeled coupled causal network.

8. The method according to claim 7, characterized in that, The process involves using the cross-stage coupling dominant causal chain set as the core input, incorporating the continuous sampling records of chip manufacturing discreteness, correcting the correlation relationships or state parameters of relevant transmission path nodes in the transmission causal chain based on manufacturing discrete information, performing mutual feedback calibration in the first direction, and generating a first-calibrated causal chain set, including: Extract the transmission path node information of each of the transmission causal chains in the cross-stage coupling dominant causal chain set, record the position, target and relationship of each transmission path node, and generate the transmission path node set of the transmission causal chain; Extract manufacturing discrete information from the continuous sampling records of the chip manufacturing discreteness, record the source, scope of influence and mode of action of each manufacturing discrete information, associate each manufacturing discrete information with the conduction path node of the corresponding chip-substrate-package coupling node, and generate the association result between discrete information and node; Based on the association results between the discrete information and the nodes, the association information or state parameters of the corresponding transmission path nodes are corrected using the manufacturing discrete information, thereby generating a corrected set of causal chain nodes. The propagation paths of the injected causal chain node set are reconnected. The updated propagation path nodes are connected according to the temporal order of the original propagation causal chain to ensure the temporal continuity of the propagation path and generate a reconnected causal chain set. The reconnected causal chain set is subjected to mutual feedback calibration in the first direction to adjust the correlation strength of the transmission path so that the transmission path of the causal chain is consistent with the influence of the manufacturing discrete information, thereby generating a first calibrated causal chain set. The calibration results are associated with the first calibrated causal chain set. A manufacturing discrete calibration-specific identifier is added to each of the first calibrated causal chains to generate a causal chain set with calibration tags. The calibration tags of the causal chain set with calibration tags are verified to generate a verified causal chain set with calibration tags.

9. The method according to claim 8, characterized in that, The process of using the manufacturing discrete information to correct the association information or state parameters of the corresponding transmission path nodes based on the association results of the discrete information and nodes, and generating a corrected causal chain node set, includes: Extract the manufacturing discrete information corresponding to each transmission path node in the association result of the discrete information and the node, extract and summarize the source, scope of influence and mode of action of each manufacturing discrete information, and generate a set of correspondence between nodes and discrete information containing the complete attributes of the manufacturing discrete information; Based on the set of correspondences between nodes and discrete information, each piece of manufacturing discrete information is added to the attribute set of the corresponding transmission path node as an influencing factor, and the node's association information or state parameters are updated accordingly to generate updated node association information. The integrated node association information is updated to the corresponding transmission path nodes of the transmission causal chain, replacing the original node association information, ensuring that the node information accurately reflects the impact of the manufacturing discrete information, and generating updated transmission path nodes; All the updated transmission path nodes are grouped according to their respective transmission causal chains, and the updated transmission path nodes under the same transmission causal chain are integrated into the same set to generate an updated node set corresponding to each transmission causal chain. Perform a node integrity check on the set of update nodes corresponding to each of the aforementioned causal chains, remove nodes with missing information or incorrect associations, ensure the integrity and accuracy of the set of update nodes, and generate a complete set of update nodes; Summarize the complete set of update nodes corresponding to all the aforementioned causal chains to generate the injected causal chain node set; The injected causal chain node set is verified to check whether the node information is completely consistent with the manufacturing discrete information, and a causal chain node set with verification identifier is generated. Add an injection identifier to the set of causal chain nodes with verification identifiers, and add an injection-specific identifier to each of the updated propagation path nodes to generate a set of causal chain nodes with injection identifiers.

10. The method according to claim 1, characterized in that, The process involves driving a thermal coupling evolution model through the mutually fed-back corrected coupled conduction causal network, performing multi-stage dynamic deduction of photoelectric and thermal conduction, generating a coupled-driven thermal evolution probability map and a dynamically unstable critical region that is updated in real time as the deduction progresses, including: The mutually fed-back modified coupled conduction causal network is input into the thermally coupled evolution model, so that the deduction logic of the thermally coupled evolution model is deeply associated with the conduction path of the mutually fed-back modified coupled conduction causal network, ensuring that the deduction process follows the logic of the mutually fed-back modified coupled conduction causal network, and generating a set of bound deduction logic. Based on the bound deduction logic set, the dynamic deduction of the multi-stage photoelectric thermal conduction is performed to simulate the complete conduction process of the photoelectric thermal effect in each stage of the thermal runaway process, record the deduction details of each stage of the thermal runaway process, and generate a multi-stage deduction result sequence. The multi-stage simulation result sequence is integrated into a graph, and the simulation results of each stage of the thermal runaway process are arranged in chronological order to generate a coupled-driven thermal operation evolution probability graph. The instability region is identified in the coupled-driven thermal operation evolution probability map, potential instability regions appearing in the simulation process are tracked in real time, the location and evolution trend of the instability region are recorded, and a dynamic instability critical domain is generated that is updated in real time with the simulation process. The coupled thermal operation evolution probability map and the dynamic instability critical domain are associated and stored, and a corresponding inference time series mark is added to each inference result to generate a set of inference results with inference time series marks. The set of inference results with inference time sequence markers is verified to check the logical consistency and accuracy of the coupled-driven thermal operation evolution probability map and the dynamic instability critical domain, to ensure the reliability of the inference results, and to generate a set of inference results with verification markers. The storage identifiers of the inference result set with verification identifiers are standardized to ensure that the storage identifier format of each inference result is consistent, thereby generating a set of inference results with standardized identifiers.

11. An early warning device, characterized in that, include: The data acquisition module is used to synchronously start the MiniLED display multi-scale sensor node to collect continuous sampling records of chip junction temperature fluctuation, electro-optical conversion efficiency decay, and substrate thermal conduction differences throughout the entire operation cycle. Through the coupling conduction link marking logic built into the sensor node, the corresponding chip-substrate-package coupling node conduction link mark is bound to each continuous sampling record to obtain a multi-scale optoelectronic thermal time series dataset anchored to the conduction link. The causal mining module is used to mine the conduction causal chain of photoelectric and thermal effects between different stages of thermal runaway based on the multi-scale photoelectric and thermal time series dataset of the anchored conduction link and the causal convolutional network. It also uses reverse causal verification logic to remove pseudo-associative sampling records and obtain a set of cross-stage coupled dominant causal chains. The mutual feedback calibration module is used to take the cross-stage coupling dominant causal chain set as the core input, introduce continuous sampling records of chip manufacturing discreteness and environmental thermal disturbance for bidirectional mutual feedback calibration, and remove the conduction deviation causal chain generated during the calibration process through the causal chain deviation conduction identification logic to obtain the coupled conduction causal network after mutual feedback correction. The dynamic simulation module is used to drive the thermal coupling evolution model through the mutually fed-back corrected coupled conduction causal network, perform dynamic simulation of multi-stage photoelectric thermal conduction, and generate a coupled-driven thermal operation evolution probability map and a dynamic instability critical domain that is updated in real time with the simulation process. The runaway early warning module is used to perform conduction causal chain matching and early warning judgment based on the thermal operation evolution probability map and dynamic instability critical domain driven by the coupling, combined with the real-time sampling records of the MiniLED display screen, to complete the thermal runaway early warning of the MiniLED display screen.

Images