Belt conveyor abnormality identification method based on industrial audio scene

Abstract

The application provides a belt conveyor abnormality identification method based on an industrial audio scene, and relates to the technical field of data processing.The method comprises the following steps: dividing industrial audio data into data units according to a fixed time span, extracting data change content of time units at adjacent time positions, identifying instantaneous change and continuous change of audio, analyzing distribution of audio change representation values on a time axis, calculating concentration degree of audio change in a fixed time range, obtaining change aggregation quantity, determining a fixed time range with a change aggregation quantity reaching a preset distribution threshold as an aggregation section, splitting the aggregation section into each time position, calculating change contribution values of each time position in a change aggregation process, and obtaining section contribution quantity.When the section contribution quantity of a certain time position meets a preset abnormality threshold, it is determined that the industrial audio at the time position has an abnormal state.The application can identify audio abnormal change of a single supporting roller under the condition of multi-supporting roller audio superposition.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for identifying abnormalities in belt conveyors based on industrial audio scenarios. Background Technology

[0002] In the field of mining material handling, the detection of belt conveyor operation status typically involves analyzing and processing audio signals generated during belt conveyor operation based on industrial audio scenarios. Industrial microphones or bone conduction stethoscopes are usually deployed near idler supports or drums along the belt conveyor to collect structural vibration sounds or airborne sounds generated during belt operation. The collected industrial audio signals are then transmitted to edge computing devices or a host platform for audio signal analysis. By extracting the temporal variation characteristics of the audio signals and comparing them with pre-established normal operation audio samples, it can be determined whether the belt conveyor is in an abnormal operating state.

[0003] In existing technologies, industrial audio acquisition typically covers multiple idlers within the same belt segment. When an idler experiences early wear or slight jamming, its abnormal audio signal may be simultaneously acquired and superimposed with the operating audio signals of several adjacent, normal idlers. For example, in the middle section of a long-distance conveyor belt in a mine, multiple idlers operate at high speed simultaneously. The abnormal audio changes caused by insufficient bearing lubrication of a single idler constitute a relatively small proportion of the overall industrial audio signal and may be easily masked by the background operating audio changes of other idlers. Consequently, when judging based on the overall audio signal change characteristics, it may not accurately reflect the actual operating status of the abnormal idler, affecting the effective identification of early idler anomalies. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying abnormalities in belt conveyors based on industrial audio scenarios, aiming to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] A method for identifying anomalies in belt conveyors based on industrial audio scenarios, the method comprising:

[0007] Acquire industrial audio data and divide it into data units according to a fixed time span. Determine the time position corresponding to each data unit to obtain time-structured audio data.

[0008] Based on the time-structured audio data, extract the data change content of time units at adjacent time positions, and convert the change content into the change state of adjacent time positions to obtain the audio change representation value;

[0009] By associating and arranging several consecutive audio change characterization values ​​in chronological order, the instantaneous and continuous changes in audio are identified, resulting in an audio change sequence.

[0010] Based on the audio change sequence, analyze the distribution of audio change characteristics on the time axis, calculate the degree of concentration of audio changes within a fixed time range, and obtain the change clustering amount;

[0011] By defining the fixed time range in which the amount of change in aggregation reaches a preset distribution threshold as the aggregation segment, candidate segment data is obtained.

[0012] Based on the candidate segment data, the clustered segments are divided into various time positions, and the contribution value of each time position in the change clustering process is calculated to obtain the segment contribution.

[0013] When the contribution of a segment at a certain time location meets the preset abnormal threshold, it is determined that there is an abnormal state in the industrial audio at that time location.

[0014] Furthermore, industrial audio data is acquired and divided into data units according to a fixed time span. The time position corresponding to each data unit is determined to obtain time-structured audio data, including:

[0015] Based on the acquisition sequence of industrial audio data, the start and end acquisition positions of the industrial audio data on the time axis are determined to obtain the time reference range;

[0016] Based on the time base range, multiple adjacent and continuous time segment intervals are generated on the time axis according to a fixed time span, and the data coverage of each time segment interval is determined to obtain a set of time segment intervals;

[0017] Based on the time segmentation interval set, extract the data segments corresponding to each time segmentation interval, and bind the data segments with the corresponding time segmentation intervals to form data units to obtain a data unit set;

[0018] Based on the data unit set, time position identifiers are assigned to each data unit according to its position on the time axis, and the data units and time position identifiers are encapsulated to obtain time-structured audio data.

[0019] Furthermore, based on the time-structured audio data, the data changes of time units at adjacent time positions are extracted, and these changes are converted into changes at adjacent time positions to obtain audio change representation values, including:

[0020] Based on the time position identifier of each data unit in the time structure audio data, adjacent data units are paired according to the order of their time positions to obtain a set of adjacent time unit pairs;

[0021] Based on the set of adjacent time units, the previous time unit in the adjacent time unit pair is taken as the reference time unit, and the reference time unit is used as the reference object for change extraction to obtain the reference time unit set;

[0022] Based on the set of reference time units, the reference time units are compared one by one with their corresponding next time units to extract the data differences between them and obtain the set of data change content.

[0023] Based on the set of data changes, the correlation between the data changes and adjacent time positions is identified, and the correlation is mapped to the state expression item of the change state to obtain the audio change representation value.

[0024] Furthermore, by associating and arranging several consecutive audio change characterization values ​​in chronological order, instantaneous and continuous changes in audio are identified, resulting in an audio change sequence, including:

[0025] Based on the audio change representation values, audio change representation values ​​with the same change state in adjacent time positions are continuously associated to obtain a set of change continuation relationships;

[0026] Based on the set of change continuation relationships, the number of time positions in which each change state appears consecutively on the time axis is counted to obtain the set of change continuation lengths;

[0027] Based on the set of change duration lengths, change states whose change duration lengths meet the preset duration threshold are identified as continuous change states, and change states whose change duration lengths do not meet the preset duration threshold are identified as instantaneous change states, thus obtaining a set of change nature identifiers;

[0028] Based on the set of change nature identifiers, the change state and change nature identifiers at each time position are encapsulated and sorted according to the order of the time position identifiers to obtain the audio change sequence.

[0029] Furthermore, based on the audio change sequence, the distribution of audio change characteristics on the time axis is analyzed, the degree of concentration of audio changes within a fixed time range is calculated, and the change clustering is obtained, including:

[0030] Based on the audio change sequence, count the number of time positions identified as continuously changing states, calculate the relative degree of continuous change within a fixed time range, and obtain the continuous occupancy item;

[0031] By determining the time positions of the first and last occurrences of the continuous change, the proportion of the time span covered by the continuous change within a fixed time range is calculated, thus obtaining the time span term;

[0032] By identifying the continuous change segments composed of adjacent time positions, the proportion of the continuous length of each continuous change segment in the whole is used to obtain the structural concentration term;

[0033] By fusing the persistent occupancy term, the time span term, and the structural concentration term, the degree of concentration of audio changes within a fixed time range is calculated to obtain the change aggregation quantity.

[0034] Furthermore, by defining the fixed time range within which the amount of change in aggregation reaches a preset distribution threshold as the aggregation segment, candidate segment data is obtained, including:

[0035] Obtain the fixed time range corresponding to each change aggregation quantity, and establish the association relationship between the change aggregation quantity and its corresponding fixed time range to obtain the time range association set;

[0036] The amount of change clustering in each fixed time range is compared with a preset distribution threshold, and fixed time ranges whose amount of change clustering meets the preset distribution threshold condition are selected to obtain a candidate time range set.

[0037] By identifying adjacent and continuous fixed time ranges on the time axis, adjacent and continuous candidate time ranges are merged into clustered segments to obtain a set of clustered segments.

[0038] Based on the clustered segment set, record the start and end time positions of each clustered segment, and encapsulate the start and end time positions to obtain candidate segment data.

[0039] Furthermore, based on the candidate segment data, the clustered segments are divided into various time positions, and the contribution value of each time position in the change clustering process is calculated to obtain the segment contribution, including:

[0040] Based on the candidate segment data and the set of change nature identifiers, the time locations identified as continuously changing states are filtered, and the degree of continuous participation of each time location within the segment is calculated to obtain the continuous driving term.

[0041] Based on the distribution of the continuously changing state within the clustering segment, identify the continuously changing segments at each time position, calculate the structural weight of the continuously changing segment corresponding to the time position, and obtain the segment structure weight term.

[0042] Based on the audio change characterization value corresponding to the time position, calculate the proportion of the change intensity at that time position relative to the overall change intensity of the cluster segment, and obtain the change intensity proportion item;

[0043] Based on the change in the clustering amount corresponding to the clustering segment of the time location, the continuous driving term, the segment structure weight term, and the change intensity ratio term are modulated at the segment level to obtain the segment clustering coupling term.

[0044] By integrating the continuous driving term, segment structure weight term, change intensity ratio term, and segment clustering coupling term, the dominant contribution of each time position in the formation process of clustered segments is calculated, and the segment contribution is obtained.

[0045] Furthermore, when the contribution of a segment at a certain time location meets a preset anomaly threshold, it is determined that the industrial audio at that time location is in an abnormal state, including:

[0046] By extracting the segment contribution at all time locations within each cluster segment, the contribution distribution within the segment is identified, and a contribution distribution set is obtained.

[0047] Based on the contribution distribution set, the dominant time position of the contribution of the segment within the same cluster segment is identified, and the dominant time position of the contribution is determined as the candidate anomaly time position, thus obtaining the candidate anomaly position set.

[0048] Based on the candidate anomaly location set, the segment consistency of the candidate anomaly time location is checked, and the time locations that do not match the overall change characteristics of the clustered segment are eliminated to obtain the confirmed anomaly location set.

[0049] Based on the confirmed abnormal location set, the industrial audio data at the corresponding time location is associated and encapsulated with its corresponding cluster segment, and the industrial audio status at that time location is determined to be an abnormal status.

[0050] Furthermore, by extracting the segment contribution at all time points within each cluster segment, the contribution distribution within each segment is identified, resulting in a contribution distribution set, including:

[0051] Based on the candidate segment data, the time location range covered by each cluster segment is obtained, and the segment contribution corresponding to each time location is extracted to obtain the segment contribution set.

[0052] Based on the segment contribution set, the segment contribution is associated with the corresponding time position according to the time position identifier, the relationship between time position and segment contribution is determined, and a contribution mapping relationship table within the segment is obtained.

[0053] Based on the contribution mapping relationship table within the segment, the distribution of segment contribution within the cluster segment is analyzed, the relative difference between segment contribution at different time locations is identified, and the segment contribution distribution characteristics are obtained.

[0054] Based on the contribution distribution characteristics of the clusters, the data reflecting the contribution relationship at each time location within the clusters are encapsulated to determine the contribution distribution within each cluster, thus obtaining the contribution distribution set.

[0055] Furthermore, based on the candidate anomaly location set, segment consistency verification is performed on the candidate anomaly time locations, and time locations that do not match the overall change characteristics of the clustered segments are eliminated to obtain the confirmed anomaly location set, including:

[0056] Based on the candidate anomaly location set, obtain the cluster segment identifier of each candidate anomaly time location, and extract the change clustering amount and segment contribution distribution set of the cluster segment to obtain the segment reference feature set;

[0057] Based on the segment reference feature set, the segment contribution of the candidate anomaly time location is compared with the overall contribution distribution characteristics of its respective cluster segment to evaluate the consistency of the contribution of the candidate anomaly time location within the segment and obtain the consistency evaluation result.

[0058] Based on the consistency assessment results, candidate abnormal time locations that do not have a dominant relationship in the segment contribution are identified and removed to obtain the consistency screening results.

[0059] Based on the consistency screening results, the time position that passes the segment consistency check is determined as the confirmed anomaly time position, thus obtaining the confirmed anomaly position set.

[0060] The above-described solution of the present invention has at least the following beneficial effects:

[0061] This invention arranges several consecutive audio change representation values ​​in chronological order and identifies instantaneous and continuous changes in audio during the association process. The data no longer remains in a set of independent change representation values, but is organized into a sequence structure with explicit temporal order and adjacent correlation. The identification of instantaneous and continuous changes ensures that each element in the sequence carries not only the change state but also the temporal persistence attribute of the change. This serialization result is equivalent to rewriting audio changes into a state sequence with persistence attributes. This eliminates the need for subsequent long-range correlation calculations on high-dimensional acoustic vectors, and allows for logical division and statistical referencing of the persistence and intermittency of events at the sequence level. This ensures that the output of subsequent clustered segments and abnormal time positions has a clear sequence source and temporal consistency.

[0062] This invention analyzes the distribution of audio change characteristics on the time axis through audio change sequences, calculates the concentration of audio changes within a fixed time range, and obtains the change cluster quantity. The distribution pattern of changes in the sequence is further compressed into a mapping from fixed time ranges to cluster quantities, so that each fixed time range corresponds to a numerically comparable concentration, providing a direct basis for subsequent candidate segment selection. It objectively expresses whether changes show a concentrated occurrence pattern within a certain time range. As a result, subsequent steps can compare different time ranges on a unified scale, without needing to make point-by-point judgments of equal complexity at each time position. This makes whether changes form clusters within a certain period of time a calculable and threshold-comparable quantity. A bridging layer from sequence to segment is established in the data processing chain, and this bridging layer still retains the boundary information of the fixed time range, which can be used as the reference basis for segment start and end location.

[0063] This invention defines a fixed time range where the amount of change in aggregation reaches a preset distribution threshold as an aggregation segment. It separates a set of time ranges that meet the aggregation conditions from a continuous time axis, transforming the subsequent processing object from a full-time-domain audio change sequence into a finite number of aggregation segments and their covered time positions. Since the candidate segment data contains the time boundary information of the aggregation segments, subsequent steps have clear range constraints when splitting and calculating contributions within the segments. It is not necessary to repeat the same calculation at time positions outside the segments. This range constraint is an objective screening result directly generated by comparing the aggregation amount with the threshold, rather than manually selecting segments. At the data processing level, it forms a segment extraction mechanism driven by distribution conditions, which is equivalent to outputting time intervals that may contain anomalies as candidate event intervals. It establishes clear data pruning boundaries and object levels in the processing chain.

[0064] This invention divides clustered segments into time positions using candidate segment data and calculates the contribution value of each time position during the clustering process to obtain the segment contribution. The clustered segment is no longer regarded as a uniform abnormality carrier, but is further decomposed into multiple time positions, and a contribution value associated with the formation of the cluster is generated for each time position. This is equivalent to quantifying and ranking the dominance of each time position within the segment. Which time positions within the segment are more critical to the formation of the cluster structure can be reflected by the relative size of the contribution value. This eliminates the need to directly return to the original audio for repeated scanning. Instead, a threshold judgment based on the contribution value can be used to obtain abnormal candidate points with clear time position identifiers, ensuring that subsequent abnormal outputs have a dual correspondence between the segment context and the time position context.

[0065] This invention defines the output object as an abnormal state in the time position dimension, rather than a general audio segment abnormality or a whole device abnormality. This allows the output to form a closed loop with the time structure audio data constructed in the first step in the data index. The abnormality judgment condition is based on the comparison of contribution amount and threshold. Therefore, each abnormal time position corresponds to a repeatable contribution amount result, and this contribution amount comes from the contribution calculation within the candidate segment. This means that the abnormal output can be directly mapped to a specific segment on the audio time axis for subsequent alarm labeling, segment playback, event recording and secondary analysis. At the same time, since the output unit is time position, one or more abnormal time positions can be output within the same candidate segment, allowing the abnormality to be presented as a combination of discrete event points or continuous event segments, without the need to preset the abnormality form. Attached Figure Description

[0066] Figure 1 This is a flowchart of a belt conveyor anomaly identification method based on industrial audio scenarios provided by an embodiment of the present invention. Detailed Implementation

[0067] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0068] like Figure 1 As shown, embodiments of the present invention propose a method for identifying belt conveyor anomalies based on industrial audio scenarios, the method comprising:

[0069] Acquire industrial audio data and divide it into data units according to a fixed time span. Determine the time position corresponding to each data unit to obtain time-structured audio data.

[0070] Based on the time-structured audio data, extract the data change content of time units at adjacent time positions, and convert the change content into the change state of adjacent time positions to obtain the audio change representation value;

[0071] By associating and arranging several consecutive audio change characterization values ​​in chronological order, the instantaneous and continuous changes in audio are identified, resulting in an audio change sequence.

[0072] Based on the audio change sequence, analyze the distribution of audio change characteristics on the time axis, calculate the degree of concentration of audio changes within a fixed time range, and obtain the change clustering amount;

[0073] By defining the fixed time range in which the amount of change in aggregation reaches a preset distribution threshold as the aggregation segment, candidate segment data is obtained.

[0074] Based on the candidate segment data, the clustered segments are divided into various time positions, and the contribution value of each time position in the change clustering process is calculated to obtain the segment contribution.

[0075] When the contribution of a segment at a certain time location meets the preset abnormal threshold, it is determined that there is an abnormal state in the industrial audio at that time location.

[0076] In this embodiment of the invention, industrial audio data is acquired and divided into data units according to a fixed time span. The time position corresponding to each data unit is determined to obtain time-structured audio data. This transforms the originally continuous and undefined industrial audio data into a set of data objects arranged in an orderly manner on the time axis with defined boundaries, providing a unified and referable time reference for all subsequent data processing steps. Based on the time-structured audio data, the data change content of time units at adjacent time positions is extracted and converted into the change state of adjacent time positions to obtain audio change characterization values. This allows subsequent analysis to be based on the change itself, making the change results between different time positions comparable and providing standardized input for the serialization and aggregation analysis of changes. By associating and arranging several continuous audio change characterization values ​​in chronological order, instantaneous and continuous changes in audio are identified to obtain an audio change sequence. By distinguishing between instantaneous and continuous changes, subsequent steps can perform distribution analysis based on the temporal continuity characteristics of the changes, avoiding the equating of isolated changes with continuous changes.

[0077] Based on the audio change sequence, the distribution of audio change characteristics on the time axis is analyzed, and the concentration of audio changes within a fixed time range is calculated to obtain the change aggregation quantity. This provides an objective basis for subsequent threshold-based time range screening, ensuring that the determination of the anomaly candidate range is based on the change distribution characteristics rather than single-point change results. By defining the fixed time range where the change aggregation quantity reaches a preset distribution threshold as the aggregation segment, candidate segment data is obtained, achieving effective pruning of the full-time domain audio and ensuring that the anomaly identification process is always limited to the time range with change aggregation characteristics. Based on the candidate segment data, the aggregation segment is divided into various time positions, and the change contribution value of each time position in the change aggregation process is calculated to obtain the segment contribution quantity. This avoids treating the entire segment as a homogeneous anomaly object and provides data basis accurate to the time position level for subsequent anomaly judgment. When the segment contribution quantity at a certain time position meets the preset anomaly threshold, it is determined that the industrial audio at that time position is in an abnormal state, ensuring the clear location of the anomaly identification result in the time dimension and providing a direct basis for audio segment location and event analysis.

[0078] Acquiring industrial audio data specifically includes:

[0079] At the operation site of the belt conveyor, industrial microphones are installed according to the installation conditions of the idler supports, drums, and nearby structural components along the belt. The pickup end face of the industrial microphone is oriented towards the main vibration propagation path of the idler or drum, and a fixed relative position is maintained between the industrial microphone and the idler support or drum support. This allows the industrial microphone to continuously receive industrial audio signals formed by the superposition of idler rotation, bearing friction, structural component resonance, and airborne sound. After the industrial microphone is installed, its sampling parameters are set so that it continuously samples the industrial audio signals at a predetermined sampling rate. The discrete sampling points obtained from the continuous sampling are then arranged in the order of acquisition time to form an original audio data stream. The original audio data stream includes structural vibration sound components and airborne sound components related to the belt's operating state, as well as background operating sound components introduced by the simultaneous operation of multiple idlers. This allows subsequent processing to reflect the superposition relationship between abnormal idler sounds and background sounds from adjacent idlers in the same audio data stream.

[0080] After the industrial microphone outputs the raw audio data stream, the raw audio data stream is input to a preset audio acquisition interface in an edge computing device or upper platform. The audio acquisition interface controls the access of the raw audio data stream, causing it to enter the data buffer in the form of continuous frames according to a predetermined transmission protocol. In the data buffer, a time reference is established for the raw audio data stream, so that each frame of data is written with a corresponding acquisition time marker when it enters the buffer, thereby establishing a definite mapping relationship between the audio data and the time axis. Subsequently, data integrity verification is performed on the raw audio data stream in the buffer. By judging whether the time markers of adjacent frames are continuous and whether the number of sampling points meets the sampling rate requirements, it is possible to identify whether there are dropped frames, duplicate frames, or time jumps. When an abnormal transmission state is detected, the corresponding frame is marked or removed to ensure that when data units are subsequently divided according to a fixed time span, the data segments covered by each data unit have a continuous and consistent sampling density.

[0081] After data access and verification, the raw audio data stream is read from the data buffer in chronological order. Amplitude normalization is then performed on the read audio data to map the audio amplitude distribution under different acquisition time periods and background sound pressure conditions to a uniform numerical range, thereby reducing amplitude scale differences caused by sensor gain fluctuations or overall changes in ambient sound pressure. After amplitude normalization, DC component suppression is performed on the audio data. By performing mean correction on the sampling point sequence of the audio data, baseline drift of the audio data is eliminated, making the extraction of data changes between subsequent adjacent time positions more focused on dynamic components. Subsequently, anti-aliasing band-limiting processing is performed on the audio data to restrict the audio data in the frequency dimension to the target frequency band related to roller rotation, bearing friction, and structural vibration, thereby reducing the impact of non-target frequency band noise on the calculation of adjacent differences, resulting in industrial audio data.

[0082] In a preferred embodiment of the present invention, industrial audio data is acquired, and the industrial audio data is divided into data units according to a fixed time span. The time position corresponding to each data unit is determined to obtain time-structured audio data, including:

[0083] Based on the acquisition sequence of industrial audio data, the start and end acquisition positions of the industrial audio data on the time axis are determined to obtain the time reference range;

[0084] Based on the time base range, multiple adjacent and continuous time segment intervals are generated on the time axis according to a fixed time span, and the data coverage of each time segment interval is determined to obtain a set of time segment intervals;

[0085] Based on the time segmentation interval set, extract the data segments corresponding to each time segmentation interval, and bind the data segments with the corresponding time segmentation intervals to form data units to obtain a data unit set;

[0086] Based on the data unit set, time position identifiers are assigned to each data unit according to its position on the time axis, and the data units and time position identifiers are encapsulated to obtain time-structured audio data.

[0087] In this embodiment of the invention, the starting and ending acquisition positions of the industrial audio data on the time axis are determined according to the acquisition sequence of the industrial audio data, thus obtaining a time reference range. This avoids data omissions or duplicate references caused by inconsistent start and end times during subsequent time division, providing a unified time reference for the generation of subsequent fixed time spans. Based on the time reference range, multiple adjacent and continuous time segmentation intervals are generated on the time axis according to the fixed time span, and the data coverage of each time segmentation interval is determined, resulting in a set of time segmentation intervals. This ensures that any processing can reproduce the same segmentation result at the same time scale, avoiding time offset problems introduced by dynamic windows or irregular segmentation. The system segments the audio stream into intervals, extracts data segments corresponding to each time interval, and binds these data segments to their corresponding time intervals as data units, thus obtaining a data unit set. This provides a stable data input object for subsequent processing such as time position-based difference extraction and sequence arrangement, avoiding repeated slicing of the original audio stream in subsequent processing. Based on the data unit set, the system assigns time position identifiers to each data unit according to its position on the time axis, and encapsulates the data units and time position identifiers to obtain time-structured audio data. This provides a unified data foundation for subsequent construction of adjacent time unit pairs, audio change sequences, and time range statistics, giving the entire audio processing workflow a clear temporal logical link.

[0088] Specifically, based on the acquisition sequence of the industrial audio data, the start and end acquisition positions of the industrial audio data on the time axis are determined to obtain the time reference range, which includes:

[0089] After an industrial microphone or bone conduction stethoscope sensor completes a continuous acquisition, the output industrial audio data is written to a buffer or storage area in the acquisition sequence, ensuring that the industrial audio data maintains an arrangement consistent with the acquisition time sequence in the storage structure. The acquisition metadata corresponding to the industrial audio data is read, and the first frame identifier and last frame identifier in the acquisition metadata are used as boundary references for the acquisition sequence. The first frame identifier corresponds to the first segment of audio data in the acquisition sequence, and the last frame identifier corresponds to the last segment of audio data in the acquisition sequence. Based on the first frame identifier, the acquisition start time of the first segment of audio data is parsed, and this acquisition start time is determined as the starting acquisition position. The acquisition end time of the last audio data segment is obtained based on the end frame identifier and is determined as the end acquisition position. When the metadata of the industrial audio data contains a sampling rate parameter, a mapping relationship is established between the sampling rate parameter and the total number of sampling points of the industrial audio data. The acquisition sequence direction is taken by the increasing direction of the sampling point number. The start acquisition position and the end acquisition position are obtained by calculating the time corresponding to the first sampling point number and the time corresponding to the last sampling point number. The time boundary can still be determined even in the absence of an explicit timestamp. The start acquisition position and the end acquisition position are encapsulated to obtain the time reference range.

[0090] Specifically, based on the time base range, multiple adjacent and continuous time segments are generated on the time axis according to a fixed time span, and the data coverage of each time segment is determined to obtain a set of time segment intervals, which includes:

[0091] First, a preset fixed time span parameter is read and used as the window length for time axis segmentation. The starting position of the time reference range is used as the starting point of the first time segmentation interval. Multiple time segmentation intervals are generated sequentially along the time axis from the starting position, ensuring that the starting point of each time segmentation interval is the ending point of the previous one. This ensures that adjacent time segmentation intervals are connected end-to-end on the time axis, avoiding time gaps or overlaps. When generating each time segmentation interval, the starting and ending points are recorded as the time boundary of that interval. Based on the sampling rate parameter or timestamp distribution of the industrial audio data, the data coverage range of this time boundary in the industrial audio data is determined, so that the data coverage range can indicate the range of sampling point numbers falling within the time boundary. Data frame sequence number range; when industrial audio data is organized by sampling points, the start and end points of the interval are converted into sampling point sequence number ranges to determine the start and end sampling point numbers covered by the time segmentation interval, thus obtaining the data coverage range; when industrial audio data is organized by data frames, the set of data frames whose timestamps fall within the interval boundaries is identified to determine the start and end indices of the data frames covered by the time segmentation interval, thus obtaining the data coverage range; when the end point of the last time segmentation interval exceeds the end acquisition position of the time reference range, the end point of the last time segmentation interval is truncated to the end acquisition position to ensure that the generated time segmentation interval set completely covers the time reference range without exceeding the boundary; all generated time segmentation intervals and their corresponding data coverage ranges are encapsulated to obtain the time segmentation interval set.

[0092] In a preferred embodiment of the present invention, based on the time-structured audio data, the data change content of time units at adjacent time positions is extracted, and the change content is converted into the change state of adjacent time positions to obtain audio change characterization values, including:

[0093] Based on the time position identifier of each data unit in the time structure audio data, adjacent data units are paired according to the order of their time positions to obtain a set of adjacent time unit pairs;

[0094] Based on the set of adjacent time units, the previous time unit in the adjacent time unit pair is taken as the reference time unit, and the reference time unit is used as the reference object for change extraction to obtain the reference time unit set;

[0095] Based on the set of reference time units, the reference time units are compared one by one with their corresponding next time units to extract the data differences between them and obtain the set of data change content.

[0096] Based on the set of data changes, the correlation between the data changes and adjacent time positions is identified, and the correlation is mapped to the state expression item of the change state to obtain the audio change representation value.

[0097] In this embodiment of the invention, based on the time position identifiers of each data unit in the time-structured audio data, adjacent data units are paired according to their chronological order to obtain a set of adjacent time unit pairs. This provides a unified data input format for subsequent pairwise comparisons, avoiding the problem of incomparable data differences caused by misaligned time windows. Based on the set of adjacent time unit pairs, the preceding time unit in each pair is used as the reference time unit, and the reference time unit is used as the reference object for change extraction, resulting in a set of reference time units. This avoids problems such as chaotic change directions or inconsistent forward and reverse comparisons, ensuring that the change results can stably reflect the changes of subsequent times relative to preceding times, providing a unified logical basis for subsequently mapping the change results to change states. The time unit set compares the baseline time unit with its corresponding next time unit one by one, extracting the data differences between them to obtain the data change content set. This avoids misjudging changes over a long period of time as instantaneous changes, providing the original basis for subsequent change state identification. This allows change judgment to rely on the difference relationship between adjacent time units rather than the absolute state of a single time unit. Based on the data change content set, the correlation between the data change content and adjacent time positions is identified, and the correlation is mapped to the state expression item of the change state to obtain the audio change characterization value. This eliminates the dependence on specific audio numerical dimensions, enabling subsequent steps to be arranged, statistically analyzed, and aggregated within a unified change state space, laying the data foundation for change sequence construction and aggregation determination.

[0098] Specifically, based on the set of adjacent time unit pairs, the preceding time unit in each pair is taken as the reference time unit, and this reference time unit is used as the reference object for change extraction, resulting in a set of reference time units, which includes:

[0099] First, the system reads the two time position identifiers corresponding to each adjacent time unit pair and automatically determines the data unit corresponding to the preceding time position as the previous time unit and the data unit corresponding to the following time position as the next time unit based on the order of the time position identifiers. After the determination, the system copies the preceding time unit as a reference object and writes the time position identifier bound to it, the adjacent pairing number, and the associated pointer with the next time unit into the reference reference object, forming a traceable reference time unit record. In order to ensure the consistency of the reference, the system performs uniform data normalization processing on the audio data of the reference time unit, so that it is under the same comparison benchmark as the next time unit. The data normalization processing includes aligning the data length, correcting the boundary of the sampling point sequence, and normalizing the time offset within the data unit, so that the two audio data segments in the same adjacent time unit pair correspond to the same sample number and time boundary. After the normalization is completed, the system converts the preceding time unit in all adjacent time unit pairs into a reference time unit record in the same way, and summarizes and stores them according to the order of the time position identifiers to obtain the reference time unit set.

[0100] Specifically, based on the set of reference time units, each reference time unit is compared with its corresponding subsequent time unit to extract the data differences between them, resulting in a set of data change content, which includes:

[0101] First, for each reference time unit record in the reference time unit set, the corresponding next time unit is obtained based on its associated pointer. Then, the same data normalization process as the reference time unit is performed on this next time unit to ensure that they are in a completely comparable isomorphic data domain. After comparability is established, the system uses the same adjacent time unit pair as the smallest comparison unit and performs difference extraction operations between the reference time unit and the next time unit according to the sample point sequence or according to a preset analysis window. The difference extraction operation is used to generate data difference results representing the relationship between the changes in two audio data segments. Specifically, it can be implemented by: performing difference operations on sampling points with the same sequence number to form a difference sequence, or performing difference operations on statistics within the same analysis window to form a window-level difference sequence, or performing difference operations on the frequency domain amplitude of the same analysis window. The distributions are compared to form a spectral difference sequence. To ensure that the difference results stably reflect the changing trends of adjacent time units, the system performs a consistent expression processing on the difference results, compressing the difference results into a difference descriptive quantity that can characterize the degree of change. The consistent expression processing includes amplitude aggregation of the difference results, unified encoding of the difference sign direction, and smoothing constraints on difference fluctuations, thereby transforming the original difference sequence into data change content that is easy to follow up on for subsequent correlation analysis. The system encapsulates the comparison results between each reference time unit and the next time unit into a data change content record, and binds the corresponding time position identifier pair in the record, namely the preceding time position and the following time position, as well as the adjacent pairing number. Finally, all data change content records are summarized in chronological order to form a data change content set.

[0102] Specifically, based on the set of data changes, the correlation between the data changes and adjacent time positions is identified, and the correlation is mapped to a state representation item of the change state to obtain the audio change representation value, which includes:

[0103] First, based on the time location identifiers bound to each data change record, the adjacent time location relationships corresponding to the change are determined. Then, the change is bound to the adjacent relationships pointing from the preceding time location to the following time location, forming an associated object that can be used to express the temporal state. Subsequently, the system performs association identification processing on the difference description quantities in the data change records to determine the change type and level of the difference description quantity in the adjacent time location relationships. This association identification processing includes: inputting the difference description quantity into a preset set of state discrimination rules, and determining whether the difference description quantity exceeds a preset change judgment threshold and whether the difference direction encoding is present. Based on specific patterns and the stability conditions of differential fluctuations in adjacent relationships, the system generates change states corresponding to the adjacent time position relationships. These change states are represented in the form of discrete state values, used to express different state categories between adjacent time positions, such as no significant change, sudden change, or the beginning of continuous change. The system encapsulates the change states and adjacent time position identifiers to form state expression items, and further writes the differential descriptor index of the source of each state expression item to ensure that the state expression item can be traced back to the specific data change content record. Finally, the system arranges the state expression items of all adjacent time positions in chronological order to obtain a set of audio change characterization values.

[0104] In a preferred embodiment of the present invention, by associating and arranging several consecutive audio change characterization values ​​in chronological order, instantaneous and continuous changes in audio are identified to obtain an audio change sequence, including:

[0105] Based on the audio change representation values, audio change representation values ​​with the same change state in adjacent time positions are continuously associated to obtain a set of change continuation relationships;

[0106] Based on the set of change continuation relationships, the number of time positions in which each change state appears consecutively on the time axis is counted to obtain the set of change continuation lengths;

[0107] Based on the set of change duration lengths, change states whose change duration lengths meet the preset duration threshold are identified as continuous change states, and change states whose change duration lengths do not meet the preset duration threshold are identified as instantaneous change states, thus obtaining a set of change nature identifiers;

[0108] Based on the set of change nature identifiers, the change state and change nature identifiers at each time position are encapsulated and sorted according to the order of the time position identifiers to obtain the audio change sequence.

[0109] In this embodiment of the invention, based on the audio change characterization value, audio change characterization values ​​with the same change state in adjacent time positions are continuously associated to obtain a change continuation relationship set, providing a data foundation for subsequent quantitative statistics on the persistence of changes and avoiding judgment based solely on single-point changes; based on the change continuation relationship set, the number of time positions in which each change state appears consecutively on the time axis is counted to obtain a change continuation length set, which distinguishes between short-term and long-term changes based on numerical criteria, providing a unified and comparable data dimension for classifying change properties; based on the change continuation length set, change states whose change continuation length meets a preset continuation threshold are determined as continuous change states, and change states whose change continuation length does not meet the preset continuation threshold are determined as instantaneous change states, resulting in a change property identifier set, which can distinguish and process different types of changes based on change properties, avoiding the mixing of instantaneous and continuous changes in subsequent statistics and analysis; based on the change property identifier set, the change state and change property identifier at each time position are encapsulated and sorted according to the chronological relationship of the time position identifiers to obtain an audio change sequence, providing a continuous and ordered data input foundation for calculating the degree of aggregation of changes within a fixed time range.

[0110] Specifically, based on the audio change representation values, audio change representation values ​​with the same change state in adjacent time positions are continuously correlated to obtain a set of change continuation relationships, which includes:

[0111] First, using audio change representation values ​​as the input basis and the already determined time position identifiers in the time-structured audio data as the time axis sequence benchmark, an index queue is constructed in ascending order of time positions. Each index element carries the change state of that time position and the corresponding audio change representation value reference information, forming a traversable sequence of adjacent time positions. Then, adjacent traversal is performed starting from the beginning time position of the index queue. During the traversal, a consistency judgment of the change state is performed on any pair of adjacent time positions. The change state expression item corresponding to the previous time position is read as the current comparison benchmark, and the change state expression item corresponding to the next time position is read as the comparison object. An equivalence judgment is performed on the state expression items to determine whether they belong to the same change state. The equivalence judgment is based on the identifier content of the change state expression item, ensuring that even if the audio change representation value itself has numerical fluctuations, as long as the mapped change state expression items are consistent, it is determined to be the same change state, merging similar changes within consecutive time periods into the same continuation chain.

[0112] When adjacent time positions are determined to have the same change state, the system performs a continuation association operation. The continuation association operation is implemented as follows: a continuation association record is generated for the adjacent time position pair, and the record is written into the change continuation relationship set. The continuation association record includes at least the previous time position identifier, the next time position identifier, and the corresponding change state identifier, and can also carry a reference to the audio change representation value corresponding to the adjacent time position pair, so that the continuation association record describes both the temporal adjacency relationship and the state consistency relationship. After writing into the relationship set, the system updates the continuation endpoint of the current comparison chain to the next time position, so that the next round of traversal can use the aforementioned next time position as the new previous time position to continue comparing with its next time position, forming a continuous continuation chain with the same change state on the time axis. When the changes in states at adjacent time points are determined to be different, the system terminates the continuation chain of the current change state and writes the chain boundary as the end marker of a continuation segment into the change continuation relationship set. The system records the start time position, end time position, and corresponding change state identifier of the current continuation segment, storing them as a segment-level encapsulation result in the change continuation relationship set, so that subsequent steps can directly calculate the continuous length based on the segment-level boundary. At the same time, the system switches the traversal reference to the next time point, sets the change state of the next time point as the new comparison reference, and begins to establish the next possible continuation chain. When traversing to the end of the timeline, the system performs a closing encapsulation on the continuation chains that have not yet been written with the termination boundary, writing the start and end time positions and change state identifier of the last continuation chain into the change continuation relationship set.

[0113] In a preferred embodiment of the present invention, based on the audio change sequence, the distribution of audio change characteristic values ​​on the time axis is analyzed, the degree of concentration of audio changes within a fixed time range is calculated, and the change aggregation amount is obtained, including:

[0114] Based on the audio change sequence, count the number of time positions identified as continuously changing states, calculate the relative degree of continuous change within a fixed time range, and obtain the continuous occupancy item;

[0115] By determining the time positions of the first and last occurrences of the continuous change, the proportion of the time span covered by the continuous change within a fixed time range is calculated, thus obtaining the time span term;

[0116] By identifying the continuous change segments composed of adjacent time positions, the proportion of the continuous length of each continuous change segment in the whole is used to obtain the structural concentration term;

[0117] By fusing the persistent occupancy term, the time span term, and the structural concentration term, the degree of concentration of audio changes within a fixed time range is calculated to obtain the change aggregation quantity.

[0118] In this embodiment of the invention, based on the audio change sequence, the number of time positions identified as continuously changing states is counted, and the relative degree of continuous change within a fixed time range is calculated to obtain a continuous occupancy term. This reflects the quantitative result of the distribution density of continuous change within the fixed time range, providing a clear data foundation for subsequent calculations of the degree of change concentration, thereby avoiding judgments based solely on single-point or scattered changes. By determining the time positions of the first and last occurrences of continuous change, the proportion of the time span covered by continuous change within the fixed time range is calculated to obtain a time span term. This quantifies the time coverage of continuous change within the fixed time range, helping to distinguish between short-term concentrated changes and cross-time distributions. The changes provide objective evidence; by identifying continuous change segments composed of adjacent time positions, the proportion of the continuous length of each continuous change segment in the whole is used to obtain the structural concentration term, distinguishing between continuous changes that occur in a scattered manner and those that occur in a concentrated manner in the form of continuous segments. This introduces a structural dimension for judging the degree of change concentration, avoiding misjudging different distribution patterns as the same state when the quantities are similar. By fusing the continuous occupancy term, the time span term, and the structural concentration term, the degree of concentration of audio changes within a fixed time range is calculated to obtain the change aggregation quantity. This can determine whether changes within different fixed time ranges show a clear concentration trend, providing a unified, stable, and repeatable quantitative basis for the determination of candidate segments.

[0119] In a preferred embodiment of the present invention, candidate segment data is obtained by defining a fixed time range in which the amount of change in aggregation reaches a preset distribution threshold as an aggregation segment, including:

[0120] Obtain the fixed time range corresponding to each change aggregation quantity, and establish the association relationship between the change aggregation quantity and its corresponding fixed time range to obtain the time range association set;

[0121] The amount of change clustering in each fixed time range is compared with a preset distribution threshold, and fixed time ranges whose amount of change clustering meets the preset distribution threshold condition are selected to obtain a candidate time range set.

[0122] By identifying adjacent and continuous fixed time ranges on the time axis, adjacent and continuous candidate time ranges are merged into clustered segments to obtain a set of clustered segments.

[0123] Based on the clustered segment set, record the start and end time positions of each clustered segment, and encapsulate the start and end time positions to obtain candidate segment data.

[0124] In this embodiment of the invention, fixed time ranges corresponding to each change aggregation are obtained, and a correlation is established between the change aggregation and its corresponding fixed time range to obtain a time range association set. This achieves a deterministic binding between the degree of change and the time interval, providing a stable data reference basis for subsequent filtering operations. The change aggregation of each fixed time range is compared with a preset distribution threshold, and fixed time ranges whose change aggregation meets the preset distribution threshold condition are filtered to obtain a candidate time range set. This achieves the first effective pruning of the time axis, providing clear filtering results for subsequent segment-level processing. By identifying adjacent and continuous fixed time ranges on the time axis, adjacent and continuous candidate time ranges are merged into aggregation segments to obtain an aggregation segment set. This avoids splitting the same change aggregation process into multiple isolated time slices, providing continuous and complete time range objects for subsequent processing. Based on the aggregation segment set, the start and end time positions of each aggregation segment are recorded, and the start and end time positions are encapsulated to obtain candidate segment data, providing clear time boundary constraints for subsequent operations.

[0125] Specifically, this involves obtaining the fixed time range corresponding to each change aggregation quantity, and establishing a correlation between the change aggregation quantity and its corresponding fixed time range to obtain a time range association set, which includes:

[0126] The system first reads each change aggregation value sequentially from the result set storing the changes aggregation values, and simultaneously reads the fixed time range boundary information used in the calculation of that change aggregation value. The fixed time range is used to define the coverage area of ​​the change aggregation value on the time axis, and its boundaries include at least a start time position and an end time position. Both the start and end time positions correspond to time position identifiers in the time-structured audio data, ensuring that the fixed time range aligns with discrete time positions on the time axis. The system then uses this fixed time range as a key-value object and the change aggregation value as associated data bound to this key-value object, generating an associated record. This associated record retains the interval semantics of the fixed time range and the numerical semantics of the change aggregation value, ensuring that the associated record uniquely points to the time axis. The system assigns a fixed time range and the corresponding concentration result to that range. It then performs the binding process on each of the variable aggregates, writing the resulting multiple associated records into the same set-type data structure. The associated records are then organized according to the chronological order of the fixed time ranges on the time axis, presenting the time range association set as a table corresponding to the fixed time ranges and the variable aggregates. When multiple fixed time ranges overlap or contain each other on the time axis, the system still establishes independent associated records for each fixed time range, ensuring that the time range association set corresponds to the calculated result of the variable aggregates. This guarantees that subsequent filtering operations can compare and select based on fixed time ranges. Through this process, the time range association set explicitly anchors each variable aggregate to its generated time range at the data structure level.

[0127] Specifically, by identifying adjacent and consecutive fixed time ranges on the timeline, adjacent and consecutive candidate time ranges are merged into clustered segments, resulting in a set of clustered segments, which includes:

[0128] The system takes a set of candidate time ranges as input. First, it extracts the start and end times of each candidate time range and constructs an interval representation on the timeline accordingly, allowing each candidate time range to be determined in terms of its temporal order and boundary connections. The system then sorts all candidate time ranges according to their start times, ensuring the order aligns with the timeline. After sorting, the system uses the current candidate time range as the interval to be merged, reading its end time as the right boundary and the start time of the next candidate time range as its left boundary. It compares the right boundary of the current interval with the left boundary of the next interval to determine if they are adjacent and continuous. When the start time of the next interval and the end time of the current interval satisfy the continuity condition on the timeline, the system identifies them as mergeable and updates the end time of the current interval to the end time of both. The later candidate time range is selected so that the merged interval covers the union of the original two candidate time ranges, thus forming a longer continuous time interval. When multiple candidate time ranges sequentially satisfy the continuity condition on the time axis, the system iteratively executes the above update operation, gradually merging multiple adjacent and continuous candidate time ranges into the same interval to be merged, until a candidate time range that does not satisfy the continuity condition is encountered. When a case of non-continuity condition is encountered, the system encapsulates the current interval to be merged into a cluster segment, records the start and end time positions of the cluster segment as the boundary description of the cluster segment, and writes the cluster segment into the cluster segment set. Then, the next candidate time range that does not satisfy the continuity condition is used as the new interval to be merged to restart the identification and merging process. Through the above operations, the system can aggregate adjacent and continuous candidate time ranges on the time axis into several non-overlapping cluster segments, so that each cluster segment corresponds to a continuous time region composed of the union of candidate time ranges, resulting in a cluster segment set.

[0129] In a preferred embodiment of the present invention, based on candidate segment data, the clustered segment is divided into various time positions, and the contribution value of each time position in the changing clustering process is calculated to obtain the segment contribution, including:

[0130] Based on the candidate segment data and the set of change nature identifiers, the time locations identified as continuously changing states are filtered, and the degree of continuous participation of each time location within the segment is calculated to obtain the continuous driving term.

[0131] Based on the distribution of the continuously changing state within the clustering segment, identify the continuously changing segments at each time position, calculate the structural weight of the continuously changing segment corresponding to the time position, and obtain the segment structure weight term.

[0132] Based on the audio change characterization value corresponding to the time position, calculate the proportion of the change intensity at that time position relative to the overall change intensity of the cluster segment, and obtain the change intensity proportion item;

[0133] Based on the change in the clustering amount corresponding to the clustering segment of the time location, the continuous driving term, the segment structure weight term, and the change intensity ratio term are modulated at the segment level to obtain the segment clustering coupling term.

[0134] By integrating the continuous driving term, segment structure weight term, change intensity ratio term, and segment clustering coupling term, the dominant contribution of each time position in the formation process of clustered segments is calculated, and the segment contribution is obtained.

[0135] In this embodiment of the invention, based on candidate segment data and a set of change property identifiers, time positions identified as continuously changing states are filtered out. The degree of continuous participation of each time position within the segment is calculated to obtain a continuous driving term. This transforms the participation information of continuously changing states within the clustered segment in the time position dimension from the identifier layer into a comparable quantifiable term, achieving a structured characterization of continuous changes within the clustered segment. Based on the distribution of continuously changing states within the clustered segment, continuously changing segments at each time position are identified, and the structural weights of the continuously changing segments corresponding to each time position are calculated to obtain segment structure weight terms. This provides a segment structure-based distinction for subsequent contribution fusion, reflecting the organizational form of continuously changing segments within the clustered segment. Based on the audio change characterization value corresponding to the time position, the change intensity of that time position relative to the overall change intensity of the clustered segment is calculated. The proportion relationship is used to obtain the change intensity proportion term, ensuring that the contribution calculation between different clustering segments or between different time positions of the same segment has a comparable and consistent dimensional basis. Based on the change clustering amount corresponding to the clustering segment to which the time position belongs, the continuous driving term, segment structure weight term, and change intensity proportion term are modulated at the segment level to obtain the segment clustering coupling term. This avoids the inconsistency in scale between segments caused by directly using unmodulated local terms to judge contributions in segments with large differences in clustering degree. The continuous driving term, segment structure weight term, change intensity proportion term, and segment clustering coupling term are fused to calculate the dominant contribution degree of each time position in the process of clustering segment formation, and obtain the segment contribution amount. This enables anomaly localization to trace back to the corresponding clustering segment and its change clustering amount background, realizing a continuous data link from segment selection to time position localization.

[0136] In a preferred embodiment of the present invention, when the contribution of a segment at a certain time location meets a preset abnormality threshold, it is determined that the industrial audio at that time location is in an abnormal state, including:

[0137] By extracting the segment contribution at all time locations within each cluster segment, the contribution distribution within the segment is identified, and a contribution distribution set is obtained.

[0138] Based on the contribution distribution set, the dominant time position of the contribution of the segment within the same cluster segment is identified, and the dominant time position of the contribution is determined as the candidate anomaly time position, thus obtaining the candidate anomaly position set.

[0139] Based on the candidate anomaly location set, the segment consistency of the candidate anomaly time location is checked, and the time locations that do not match the overall change characteristics of the clustered segment are eliminated to obtain the confirmed anomaly location set.

[0140] Based on the confirmed abnormal location set, the industrial audio data at the corresponding time location is associated and encapsulated with its corresponding cluster segment, and the industrial audio status at that time location is determined to be an abnormal status.

[0141] In this embodiment of the invention, by extracting the segment contribution amount at all time positions within each cluster segment, the contribution distribution within the segment is identified, resulting in a contribution distribution set. This set clearly reflects the relative contribution status of different time positions within the same segment, providing a direct data foundation for subsequent identification of the dominant time position within the segment. Based on the contribution distribution set, the dominant contribution time position within the same cluster segment is identified, and this dominant contribution time position is determined as a candidate abnormal time position, resulting in a candidate abnormal position set. This ensures that the determination of the candidate abnormal time positions directly stems from their dominant role within the segment, rather than single-point threshold triggering. Based on the candidate abnormal... The location set performs segment consistency verification on candidate abnormal time locations, eliminating time locations that do not match the overall change characteristics of the clustered segment, resulting in a confirmed abnormal location set. This avoids misjudging time locations that contribute abnormally locally but are unrelated to the overall change of the segment as abnormal. Based on the confirmed abnormal location set, the industrial audio data at the corresponding time location is associated and encapsulated with its respective clustered segment, and the industrial audio status at that time location is determined as an abnormal status. This achieves a direct mapping between abnormal results and the input audio time axis, ensuring that all abnormal outputs are subject to the dual constraints of contribution-driven screening and segment consistency verification, providing a clear data index foundation for the monitoring and tracing of abnormal audio.

[0142] Specifically, based on the contribution distribution set, the dominant time positions of segment contributions within the same cluster segment are identified, and these dominant time positions are determined as candidate anomaly time positions, resulting in a candidate anomaly position set, which includes:

[0143] First, based on the candidate segment data, the start and end times of each cluster segment on the time axis are determined. Within this time range, the contribution amount of each segment corresponding to each time position in the contribution distribution set is read one by one, forming a sequential alignment sequence of time position indicators and segment contributions within the cluster segment. Then, within the same cluster segment, the segment contribution amount is used as a comparison object, and a sliding comparison is performed on the segment contribution amount according to the order of the time position indicators, establishing a local comparison relationship between the segment contribution amount at each time position and the segment contribution amount at its adjacent time positions. This yields a local difference sequence of segment contribution amounts within the cluster segment, and this local difference sequence is used to identify the rising, peak, and falling segments of the segment contribution amount on the time axis, explicitly expressing the variation pattern of the segment contribution amount within the segment. Further, within the same cluster segment, segment contribution amounts are normalized, using the maximum value of the segment contribution amount within the cluster segment as the benchmark and generating a relative contribution ratio within the segment. This isolates the dimensional differences caused by differences in the magnitude of cluster amounts between different cluster segments, thus preventing subsequent candidate anomalies from being identified. The selection of time locations relies solely on the relative dominance within a segment. Subsequently, the dominant contribution time location is identified based on the relative contribution ratio within the segment. Specifically, by scanning all time locations within the segment, the range of time locations where the relative contribution ratio reaches its extreme value within the segment and satisfies the continuity window condition near the extreme value is determined. The center time location or the earliest time location to reach the extreme value within this range is identified as the dominant contribution time location, ensuring that the dominant contribution time location corresponds to the peak core in the segment's contribution distribution pattern. When the segment's contribution distribution exhibits a multi-peak structure, the dominance of each peak is determined based on the relative contribution ratio. Specifically, by comparing the relative contribution ratio corresponding to each peak and the attenuation slope on both sides of the peak, the peak with stronger dominance is identified, and its corresponding time location is selected as the dominant contribution time location. This avoids redundancy in candidate locations caused by secondary peaks within the same clustered segment. After completing the identification of the dominant contribution time location, the time location identifier of the dominant contribution time location is associated and encapsulated with its corresponding segment identifier, and this time location identifier is written into the candidate abnormal location set.

[0144] Specifically, based on the confirmed abnormal location set, the industrial audio data at the corresponding time location is associated and encapsulated with its corresponding aggregation segment, and the industrial audio status at that time location is determined to be an abnormal status, including:

[0145] First, based on the confirmed anomaly location set, the time location identifier of each confirmed anomaly time location is read, and the data unit corresponding to the time location identifier is retrieved from the time structure audio data according to the time location identifier. This allows the confirmed anomaly time location to be located within a fixed time span of industrial audio data segments, thus obtaining the anomaly time segment. Next, based on the cluster segment identifier carried in the confirmed anomaly time location, the cluster segment boundary information corresponding to the cluster segment identifier is retrieved from the candidate segment data to obtain the start and end time positions of the cluster segment. All data units covered by the cluster segment boundary are extracted from the time structure audio data to form a cluster segment audio set, ensuring that the anomaly time segment and its context segment audio are on the same time reference. Further, the anomaly time segment and the cluster segment audio set are associated and encapsulated. Specifically, the anomaly time segment is written as a core segment field into the encapsulation structure, and the start and end time positions of the cluster segment audio set, the contribution distribution feature index within the segment, and the corresponding data unit are also included. The segment contribution is written as a context field into the encapsulation structure, so that the abnormal output carries a complete data chain of abnormal segments, segment context, and contribution basis. After encapsulation, according to the definition of the confirmed abnormal location set, the industrial audio state corresponding to the abnormal time position is set as an abnormal state, and an abnormal state identifier record is generated. The abnormal state identifier record includes at least an abnormal time position identifier, an abnormal state type field, and an abnormal judgment basis field. The abnormal judgment basis field writes the comparison result of the segment contribution at that time position and the abnormal threshold, so that the abnormal state is not an isolated label but is bound to the threshold judgment link. Finally, the above encapsulation structure and abnormal state identifier record are output to the upper platform or edge storage module and registered on the time axis according to the time position identifier, so that the abnormal time segment can be replayed or further analyzed based on the time position identifier. At the same time, the segment audio context before and after the abnormality can be traced based on the cluster segment boundary, thus completing the determination and structured output of the abnormal state of the industrial audio corresponding to the confirmed abnormal location set.

[0146] In a preferred embodiment of the present invention, by extracting the segment contribution at all time positions within each aggregation segment, the contribution distribution within the segment is identified, resulting in a contribution distribution set, including:

[0147] Based on the candidate segment data, the time location range covered by each cluster segment is obtained, and the segment contribution corresponding to each time location is extracted to obtain the segment contribution set.

[0148] Based on the segment contribution set, the segment contribution is associated with the corresponding time position according to the time position identifier, the relationship between time position and segment contribution is determined, and a contribution mapping relationship table within the segment is obtained.

[0149] Based on the contribution mapping relationship table within the segment, the distribution of segment contribution within the cluster segment is analyzed, the relative difference between segment contribution at different time locations is identified, and the segment contribution distribution characteristics are obtained.

[0150] Based on the contribution distribution characteristics of the clusters, the data reflecting the contribution relationship at each time location within the clusters are encapsulated to determine the contribution distribution within each cluster, thus obtaining the contribution distribution set.

[0151] In this embodiment of the invention, based on candidate segment data, the time location range covered by each cluster segment is obtained, and the segment contribution corresponding to each time location is extracted to obtain a segment contribution set. This provides a unified data input format for subsequent processing, avoiding interference from time locations outside the segment on anomaly detection within the segment. Based on the segment contribution set, the segment contribution is associated with the corresponding time location according to the time location identifier, determining the relationship between time location and segment contribution, and obtaining a contribution mapping table within the segment. This provides a direct data foundation for subsequent distribution state analysis, enabling the system to compare contribution differences at different time locations in the time dimension and supporting contribution distribution judgment based on time location. The contribution relationship table analyzes the distribution of segment contributions within clustered segments, identifies the relative differences in segment contributions at different time points, and obtains the segment contribution distribution characteristics. This objectively reflects whether the contributions within a segment are evenly distributed or concentrated, providing a quantitative basis for subsequent identification of the dominant time position within the segment. Based on the segment contribution distribution characteristics, the data reflecting the contribution relationships at each time point within the clustered segments are encapsulated to determine the contribution distribution within each clustered segment, resulting in a contribution distribution set. This provides a direct basis for subsequent anomaly consistency judgment, ensuring that the internal contribution relationships of each clustered segment can be individually referenced and verified, enhancing the interpretability and traceability of the overall anomaly identification process.

[0152] Specifically, based on the contribution mapping table within the segment, the distribution of segment contributions within the clustered segment is analyzed, the relative differences in segment contributions at different time locations are identified, and the segment contribution distribution characteristics are obtained, including:

[0153] First, each cluster segment is treated as an independent analysis object. All time-location markers covered by the cluster segment are sequentially arranged according to chronological order, forming a sequence structure in the mapping table consistent with the time axis. In this sequence structure, each record is considered a correspondence between a time-location marker and the segment's contribution. All segment contributions within the cluster segment are read to obtain the segment-specific sample set of contribution values, and the time-location index corresponding to each sample is retained for subsequent location and reference of differences. Based on this, a baseline statistic for the segment contribution value is calculated for the segment-specific sample set. This baseline statistic includes the number of samples in the segment-specific sample set, the sample ranking result, the representative value of the center position, and the representative value of the dispersion, providing a unified reference for subsequent difference identification. Subsequently, the segment-specific sample set is sorted from largest to smallest segment contribution value, and the numerical interval relationship between samples from the preceding and following segments is extracted based on the ranking result, forming a difference characterization to determine whether there is a clear stratification of contribution values. The use of heterogeneous interval data allows for the representation of time positions within a segment where significant deviations in contribution can be identified through interval relationships. Pairwise comparisons of segment contributions at adjacent time positions are performed along the time axis to obtain adjacent difference sequences. These adjacent difference sequences are then jointly analyzed with the aforementioned sorted difference interval data to identify the continuous pattern of contribution differences on the time axis, distinguishing between whether high-contribution time positions form continuous segments and whether high contributions exhibit isolated peaks. During the joint analysis, high-contribution samples are aggregated according to time position identifiers to generate a high-contribution candidate set. The distribution span, number of intervals, and adjacent continuity relationships of this candidate set within the segment serve as a structured description of the difference relationships within the segment, yielding segment contribution distribution characteristics. These segment contribution distribution characteristics include at least: discrete feature terms characterizing the overall dispersion of contributions within the segment; deviation feature terms characterizing whether significant deviations in contributions exist within the segment; and continuity feature terms characterizing whether high-contribution time positions form continuous segments on the time axis.

[0154] Specifically, based on the contribution distribution characteristics of each cluster segment, the data reflecting the contribution relationship at each time location within the cluster segment are encapsulated to determine the contribution distribution within each cluster segment, resulting in a contribution distribution set, which specifically includes:

[0155] First, a binding relationship is established between the segment contribution distribution feature and the segment identifier of its source cluster segment, so that any subsequent call can index the corresponding distribution result through the segment identifier. Then, a structured encapsulation process is performed on the segment contribution distribution feature. This encapsulation process includes writing discrete feature items, deviation feature items, continuous feature items, and location reference items into the segment distribution record according to predetermined fields, and storing the segment distribution record together with the start and end time positions of the cluster segment, so that the segment distribution record has clear time boundaries while maintaining the ability to describe the distribution pattern. During the encapsulation process, to ensure that the contribution relationship within the segment can be verified, the key intermediate results used to generate the distribution feature are encapsulated using references. These key intermediate results include at least an index reference to the contribution mapping relationship table within the segment, an index reference for sorting the segment contribution amount, and a reference to high contribution amounts. The temporal location reference of the candidate set allows the segment distribution records to not only provide results but also retain the corresponding data entry points. After single-segment encapsulation, multiple segment distribution records are aggregated and summarized according to the order of clustered segments in the candidate segment data to form a contribution distribution set. An index structure of segment identifiers and segment distribution records is established at the contribution distribution set level, enabling the contribution distribution set to support fast retrieval by segment, retrieval by time range, and retrieval by feature item. Finally, the contribution distribution set undergoes consistency verification processing. This consistency verification processing includes verifying whether the time boundaries in each segment distribution record are consistent with the candidate segment data, verifying whether the temporal location identifiers pointed to by the positioning reference items all fall within the coverage range of the corresponding clustered segment, and verifying whether the calculation inputs of discrete feature items and deviation feature items all come from the intra-segment contribution mapping relationship table of the same clustered segment.

[0156] In a preferred embodiment of the present invention, the candidate anomaly time locations are checked for segment consistency based on the candidate anomaly location set, and time locations that do not match the overall change characteristics of the clustered segments are eliminated to obtain a confirmed anomaly location set, including:

[0157] Based on the candidate anomaly location set, obtain the cluster segment identifier of each candidate anomaly time location, and extract the change clustering amount and segment contribution distribution set of the cluster segment to obtain the segment reference feature set;

[0158] Based on the segment reference feature set, the segment contribution of the candidate anomaly time location is compared with the overall contribution distribution characteristics of its respective cluster segment to evaluate the consistency of the contribution of the candidate anomaly time location within the segment and obtain the consistency evaluation result.

[0159] Based on the consistency assessment results, candidate abnormal time locations that do not have a dominant relationship in the segment contribution are identified and removed to obtain the consistency screening results.

[0160] Based on the consistency screening results, the time position that passes the segment consistency check is determined as the confirmed anomaly time position, thus obtaining the confirmed anomaly position set.

[0161] In this embodiment of the invention, based on the candidate anomaly location set, the cluster segment identifier of each candidate anomaly time location is obtained, and the change clustering amount and segment contribution distribution set of the cluster segment are extracted to obtain a segment reference feature set, providing a stable and reusable segment-level reference for subsequent consistency assessment. Based on the segment reference feature set, the segment contribution amount of the candidate anomaly time location is compared with the overall contribution distribution characteristics of its respective cluster segment to assess the degree of consistency of the candidate anomaly time location's contribution within the segment, obtaining a consistency assessment result. This avoids misjudging isolated high values ​​unrelated to the overall change structure of the segment as anomalies, improving the determination of anomaly time locations and the correlation between cluster segment change patterns. The structural consistency between segments is assessed. Based on the consistency assessment results, candidate abnormal time positions that do not have a dominant relationship in the segment contribution are identified and removed to obtain the consistency screening results. This reduces the probability of misjudgment of anomalies caused by local fluctuations or occasional changes, and ensures the correspondence between the abnormal output and the dominant mechanism of segment changes. Based on the consistency screening results, the time positions that pass the segment consistency verification are determined as confirmed abnormal time positions, resulting in a confirmed abnormal position set. This achieves a reliable transition from candidate anomalies to confirmed anomalies, ensuring that the output abnormal time positions have a clear and stable reference relationship in both the time structure audio data and the clustered segment structure.

[0162] Specifically, based on the segment reference feature set, the segment contribution of the candidate anomaly time location is compared with the overall contribution distribution characteristics of its corresponding cluster segment to assess the consistency of the candidate anomaly time location's contribution within the segment, thus obtaining a consistency assessment result, which includes:

[0163] First, based on the clustering segment identifier carried by the candidate anomaly time location, the change clustering amount and segment contribution distribution set of the corresponding clustering segment are retrieved from the segment reference feature set, and the time location range covered by the clustering segment is used as the boundary of the comparison domain. Then, the segment contribution amount of the candidate anomaly time location is read, and the segment contribution amount sequence of all time locations within the clustering segment is extracted within the comparison domain boundary to form a segment contribution reference sequence for distribution calculation. Next, the overall contribution distribution characteristics are calculated based on the segment contribution reference sequence. The overall contribution distribution characteristics include at least the relative ordering relationship within the segments of the segment contribution reference sequence, the concentration interval of the contribution distribution, and the dispersion of the contribution distribution, used to characterize the structural form of the contribution amount within the clustering segment on the time axis. After obtaining the overall contribution distribution characteristics, the segment contribution amount of the candidate anomaly time location is mapped to the overall contribution distribution characteristics. In the overall contribution distribution characteristics, the relative ranking of the candidate anomaly time location within the segment and its distance relationship with the contribution concentration interval are determined, and the contribution consistency degree is calculated accordingly. The contribution consistency degree is used to reflect whether the contribution performance of the candidate anomaly time location is within the dominant contribution structure of the cluster segment. Specifically, it is reflected in whether the segment contribution of the candidate anomaly time location falls into the dominant interval of the contribution distribution, whether it is consistent with the concentration trend of the contribution distribution, and whether it forms a continuous or adjacent structural association with other high contribution time locations within the segment. Finally, the contribution consistency degree is compared with the preset consistency judgment conditions, and the consistency evaluation result is output. The consistency evaluation result is used to indicate whether the candidate anomaly time location has structural consistency with the overall contribution distribution characteristics within its respective cluster segment, providing direct data basis for subsequent processing.

[0164] Based on the consistency assessment results, candidate anomalous time locations that do not have a dominant relationship in the segment contribution are identified and removed to obtain the consistency screening results, which specifically include:

[0165] First, the candidate anomaly time locations in the candidate anomaly location set are associated and encapsulated with their corresponding consistency assessment results to form candidate anomaly consistency comparison data. This data is then grouped by cluster segment identifier, ensuring that candidate anomaly time locations within the same cluster segment are in the same filtering domain. Subsequently, within each filtering domain, candidate anomaly time locations deemed not to meet the consistency criteria are extracted based on the consistency assessment results and marked as non-dominant time locations. The criteria for determining non-dominant time locations include at least the fact that the segment contribution of the candidate anomaly time location does not enter the dominant interval in the relative ranking within the segment, and that the candidate anomaly time location and its contribution... If the distance between clustered intervals exceeds a preset distance condition and the candidate anomaly time location does not form a continuous or adjacent structural association with the high-contribution structure within the segment, the dominant relationship is objectively limited by the distribution structure characteristics within the segment. After marking the non-dominant time locations, the marked candidate anomaly time locations are removed from the candidate anomaly location cluster to obtain the consistency screening results. During the removal process, the clustering segment identifier, segment contribution, and consistency assessment results corresponding to the candidate anomaly time location are retained as traceability data, so that the consistency screening results can simultaneously represent the screening basis of the retained time locations and the removal reason of the removed time locations. Finally, the screening domain outputs of each clustering segment are merged to form the consistency screening results.

[0166] Based on the consistency screening results, the time position that passes the segment consistency check is determined as the confirmed anomaly time position, resulting in a confirmed anomaly position set, which specifically includes:

[0167] First, the time locations in the consistency screening results are validated for validity. This validation includes at least verifying whether the time location still falls within the time range covered by its original cluster segment, whether the time location has a complete record of segment contribution, and whether a corresponding segment reference feature set association item exists. This ensures that the confirmed output time locations have a complete data reference chain. Subsequently, the time locations that pass the validity validation are used as a candidate set for confirming abnormal time locations. These are then repackaged according to the cluster segment identifier to form a correspondence structure between cluster segments and confirmed abnormal time locations, facilitating subsequent anomaly analysis by segment. The results are then output; next, based on the aforementioned correspondence structure, each confirmed anomaly time location is appended with its corresponding segment contribution and its relative structural position identifier within the overall contribution distribution characteristics, so that the confirmed anomaly time location not only provides a time index but also carries confirmation criteria associated with the contribution structure within the segment; finally, the encapsulated set of confirmed anomaly time locations is output as a set of confirmed anomaly locations, enabling the set of confirmed anomaly locations to directly locate the industrial audio segment corresponding to the anomaly on the time axis, and to trace back to the change in the aggregation amount and segment contribution distribution characteristics of the corresponding aggregation segment, thereby realizing the deterministic correlation between the confirmed output of the anomaly time location and the segment-level structural background.

[0168] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying belt conveyor anomalies based on industrial audio scenarios, characterized in that, The method includes: Acquire industrial audio data and divide it into data units according to a fixed time span. Determine the time position corresponding to each data unit to obtain time-structured audio data. Based on the time-structured audio data, extract the data change content of time units at adjacent time positions, and convert the change content into the change state of adjacent time positions to obtain the audio change representation value; By associating and arranging several consecutive audio change characterization values ​​in chronological order, the instantaneous and continuous changes in audio are identified, resulting in an audio change sequence. Based on the audio change sequence, the distribution of audio change characteristics on the time axis is analyzed, the degree of concentration of audio changes within a fixed time range is calculated, and the change clustering is obtained, including: Based on the audio change sequence, count the number of time positions identified as continuously changing states, calculate the relative degree of continuous change within a fixed time range, and obtain the continuous occupancy item; By determining the time positions of the first and last occurrences of the continuous change, the proportion of the time span covered by the continuous change within a fixed time range is calculated, thus obtaining the time span term; By identifying the continuous change segments composed of adjacent time positions, the proportion of the continuous length of each continuous change segment in the whole is used to obtain the structural concentration term; By fusing the persistent occupancy term, the time span term, and the structural concentration term, the degree of concentration of audio changes within a fixed time range is calculated to obtain the change aggregation quantity; By defining the fixed time range in which the amount of change in aggregation reaches a preset distribution threshold as the aggregation segment, candidate segment data is obtained. Based on the candidate segment data, the clustered segments are divided into various time positions, and the contribution value of each time position in the change clustering process is calculated to obtain the segment contribution. When the contribution of a segment at a certain time location meets the preset abnormal threshold, it is determined that there is an abnormal state in the industrial audio at that time location.

2. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 1, characterized in that, Acquire industrial audio data and divide it into data units according to a fixed time span. Determine the time position corresponding to each data unit to obtain time-structured audio data, including: Based on the acquisition sequence of industrial audio data, the start and end acquisition positions of the industrial audio data on the time axis are determined to obtain the time reference range; Based on the time base range, multiple adjacent and continuous time segment intervals are generated on the time axis according to a fixed time span, and the data coverage of each time segment interval is determined to obtain a set of time segment intervals; Based on the time segmentation interval set, extract the data segments corresponding to each time segmentation interval, and bind the data segments with the corresponding time segmentation intervals to form data units to obtain a data unit set; Based on the data unit set, time position identifiers are assigned to each data unit according to its position on the time axis, and the data units and time position identifiers are encapsulated to obtain time-structured audio data.

3. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 2, characterized in that, Based on the time-structured audio data, the data changes of time units at adjacent time positions are extracted, and the changes are converted into changes at adjacent time positions to obtain audio change representation values, including: Based on the time position identifier of each data unit in the time structure audio data, adjacent data units are paired according to the order of their time positions to obtain a set of adjacent time unit pairs; Based on the set of adjacent time units, the previous time unit in the adjacent time unit pair is taken as the reference time unit, and the reference time unit is used as the reference object for change extraction to obtain the reference time unit set; Based on the set of reference time units, the reference time units are compared one by one with their corresponding next time units to extract the data differences between them and obtain the set of data change content. Based on the set of data changes, the correlation between the data changes and adjacent time positions is identified, and the correlation is mapped to the state expression item of the change state to obtain the audio change representation value.

4. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 3, characterized in that, By associating and arranging several consecutive audio change characterization values ​​in chronological order, instantaneous and continuous changes in audio are identified, resulting in an audio change sequence, including: Based on the audio change representation values, audio change representation values ​​with the same change state in adjacent time positions are continuously associated to obtain a set of change continuation relationships; Based on the set of change continuation relationships, the number of time positions in which each change state appears consecutively on the time axis is counted to obtain the set of change continuation lengths; Based on the set of change duration lengths, change states whose change duration lengths meet the preset duration threshold are identified as continuous change states, and change states whose change duration lengths do not meet the preset duration threshold are identified as instantaneous change states, thus obtaining a set of change nature identifiers; Based on the set of change nature identifiers, the change state and change nature identifiers at each time position are encapsulated and sorted according to the order of the time position identifiers to obtain the audio change sequence.

5. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 4, characterized in that, By defining the fixed time range within which the amount of variation in clustering reaches a preset distribution threshold as the clustering segment, candidate segment data is obtained, including: Obtain the fixed time range corresponding to each change aggregation quantity, and establish the association relationship between the change aggregation quantity and its corresponding fixed time range to obtain the time range association set; The amount of change clustering in each fixed time range is compared with a preset distribution threshold, and fixed time ranges whose amount of change clustering meets the preset distribution threshold condition are selected to obtain a candidate time range set. By identifying adjacent and continuous fixed time ranges on the time axis, adjacent and continuous candidate time ranges are merged into clustered segments to obtain a set of clustered segments. Based on the clustered segment set, record the start and end time positions of each clustered segment, and encapsulate the start and end time positions to obtain candidate segment data.

6. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 5, characterized in that, Based on the candidate segment data, the clustered segments are divided into various time positions, and the contribution value of each time position in the change clustering process is calculated to obtain the segment contribution, including: Based on the candidate segment data and the set of change nature identifiers, the time locations identified as continuously changing states are filtered, and the degree of continuous participation of each time location within the segment is calculated to obtain the continuous driving term. Based on the distribution of the continuously changing state within the clustering segment, identify the continuously changing segments at each time position, calculate the structural weight of the continuously changing segment corresponding to the time position, and obtain the segment structure weight term. Based on the audio change characterization value corresponding to the time position, calculate the proportion of the change intensity at that time position relative to the overall change intensity of the cluster segment, and obtain the change intensity proportion item; Based on the change in the clustering amount corresponding to the clustering segment of the time location, the continuous driving term, the segment structure weight term, and the change intensity ratio term are modulated at the segment level to obtain the segment clustering coupling term. By integrating the continuous driving term, segment structure weight term, change intensity ratio term, and segment clustering coupling term, the dominant contribution of each time position in the formation process of clustered segments is calculated, and the segment contribution is obtained.

7. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 6, characterized in that, When the contribution of a segment at a certain time location meets a preset anomaly threshold, it is determined that there is an abnormal state in the industrial audio at that time location, including: By extracting the segment contribution at all time locations within each cluster segment, the contribution distribution within the segment is identified, and a contribution distribution set is obtained. Based on the contribution distribution set, the dominant time position of the contribution of the segment within the same cluster segment is identified, and the dominant time position of the contribution is determined as the candidate anomaly time position, thus obtaining the candidate anomaly position set. Based on the candidate anomaly location set, the segment consistency of the candidate anomaly time location is checked, and the time locations that do not match the overall change characteristics of the clustered segment are eliminated to obtain the confirmed anomaly location set. Based on the confirmed abnormal location set, the industrial audio data at the corresponding time location is associated and encapsulated with its corresponding cluster segment, and the industrial audio status at that time location is determined to be an abnormal status.

8. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 7, characterized in that, By extracting the segment contribution at all time locations within each cluster segment, the contribution distribution within each segment is identified, resulting in a contribution distribution set, including: Based on the candidate segment data, the time location range covered by each cluster segment is obtained, and the segment contribution corresponding to each time location is extracted to obtain the segment contribution set. Based on the segment contribution set, the segment contribution is associated with the corresponding time position according to the time position identifier, the relationship between time position and segment contribution is determined, and a contribution mapping relationship table within the segment is obtained. Based on the contribution mapping relationship table within the segment, the distribution of segment contribution within the cluster segment is analyzed, the relative difference between segment contribution at different time locations is identified, and the segment contribution distribution characteristics are obtained. Based on the contribution distribution characteristics of the clusters, the data reflecting the contribution relationship at each time location within the clusters are encapsulated to determine the contribution distribution within each cluster, thus obtaining the contribution distribution set.

9. The belt conveyor anomaly identification method based on industrial audio scenarios according to claim 8, characterized in that, Based on the candidate anomaly location set, segment consistency checks are performed on the candidate anomaly time locations. Time locations that do not match the overall change characteristics of the clustered segments are removed, resulting in the confirmed anomaly location set, including: Based on the candidate anomaly location set, obtain the cluster segment identifier of each candidate anomaly time location, and extract the change clustering amount and segment contribution distribution set of the cluster segment to obtain the segment reference feature set; Based on the segment reference feature set, the segment contribution of the candidate anomaly time location is compared with the overall contribution distribution characteristics of its respective cluster segment to evaluate the consistency of the contribution of the candidate anomaly time location within the segment and obtain the consistency evaluation result. Based on the consistency assessment results, candidate abnormal time locations that do not have a dominant relationship in the segment contribution are identified and removed to obtain the consistency screening results. Based on the consistency screening results, the time position that passes the segment consistency check is determined as the confirmed anomaly time position, thus obtaining the confirmed anomaly position set.

Images