A method for monitoring the state of a stacker-reclaimer based on intelligent sensors

Abstract

The application discloses a kind of based on intelligent sensor's stacker-reclaimer state monitoring method, comprising the following steps: based on intelligent sensor acquisition operating state data, generates stacker-reclaimer state sequence;Phase chain MTAD-GAT model is constructed;Stacker-reclaimer state sequence is divided into mechanism partition state sequence;Based on mechanism partition state sequence, calculate local attention relationship and cross-mechanism attention relationship, generate mechanism chain state representation;Based on stacker-reclaimer state sequence, it is divided into different working condition phase fragments, generate phase internal state representation;Based on phase internal state representation, calculate phase transition relationship, generate phase mechanism coupling state representation;Based on phase mechanism coupling state representation and mechanism partition state sequence, generate equipment state abnormal score and sensor credibility abnormal score;Stacker-reclaimer state monitoring information is constructed.The application improves abnormal positioning accuracy and state monitoring reliability.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance of bulk material handling equipment in ports and mines, and in particular to a method for monitoring the status of stacker-reclaimers based on intelligent sensors. Background Technology

[0002] Stacker-reclaimers are widely used in ports, mines, power plants, and bulk material storage yards. Their operation involves the coordinated work of multiple mechanisms, including rotation, pitching, traveling, bucket wheel reclaiming, and belt conveying. These machines operate under high loads, strong vibrations, high dust concentrations, and frequent changes in operating conditions, making them prone to bearing wear, hydraulic fluctuations, structural fatigue, sensor drift, and localized transmission abnormalities.

[0003] Existing methods for monitoring the condition of stacker-reclaimers often rely on threshold judgments for single-data types such as vibration, temperature, current, pressure, and load, or directly input multi-source sensor data into a general time-series anomaly detection model for identification. These methods typically do not adequately consider the power transmission, load transmission, and vibration propagation relationships between different mechanisms of the stacker-reclaimer, making it difficult to distinguish between the source of the anomaly and its associated responses, and easily leading to inaccurate location of the abnormal mechanism.

[0004] Meanwhile, during unloaded movement, operational engagement, stable material reclaiming, attitude adjustment, and shutdown holding, sensor data from the stacker-reclaimer undergoes significant changes depending on the phase of the operating condition. Existing methods, which analyze operational data based solely on fixed time windows, are prone to misinterpreting normal operating condition transitions as equipment malfunctions and struggle to promptly identify genuine anomalies occurring during these transitions.

[0005] Furthermore, field smart sensors are susceptible to measurement drift due to dust, impact, temperature drift, and loose installation. Existing condition monitoring methods typically treat sensor malfunctions as part of equipment malfunctions, lacking a mechanism to separate and determine sensor reliability from equipment condition abnormalities, leading to false alarms, missed alarms, and unclear troubleshooting directions.

[0006] Therefore, how to provide a method for monitoring the status of stacker-reclaimers based on intelligent sensors is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a stacker-reclaimer status monitoring method based on intelligent sensors. This invention generates a stacker-reclaimer status sequence based on operating status data, constructs a phase chain MTAD-GAT model, generates a mechanism chain status representation through a mechanism chain partitioning attention unit, generates a phase mechanism coupling status representation through a working condition phase slicing unit and a phase transition coupling unit, and generates equipment status anomaly scores and sensor reliability anomaly scores through a sensor reliability gating unit and a prediction reconstruction separation unit. This allows for the determination of abnormal mechanism partitions, abnormal sensor locations, equipment status anomaly results, and sensor reliability anomaly results, possessing advantages such as accurate anomaly location, strong working condition phase adaptability, and high sensor anomaly discrimination capability.

[0008] A stacker-reclaimer status monitoring method based on intelligent sensors according to an embodiment of the present invention includes the following steps: Based on the intelligent sensors installed on the stacker-reclaimer, the operating status data is collected, data preprocessing is performed, and a stacker-reclaimer status sequence is generated; The phase chain MTAD-GAT model is constructed, including mechanism chain partitioning attention unit, working condition phase slicing unit, phase transition coupling unit, sensor confidence gating unit and prediction reconstruction separation unit; Based on the content of each mechanism corresponding to the operating status data, the stacker-reclaimer status sequence is divided into mechanism partition status sequences; Based on the mechanism partition state sequence, mechanism transfer relations are constructed in the mechanism chain partition attention unit, local attention relations and cross-mechanism attention relations are calculated, and mechanism chain state representation is generated; Based on the stacker-reclaimer state sequence, the working condition phase slicing unit is called to determine the working condition phase, and the stacker-reclaimer state sequence is divided into different working condition phase segments to generate a state representation within the phase; Based on the phase state representation, the phase transition relationship between different working condition phase segments is calculated in the phase transition coupling unit to generate the phase mechanism coupling state representation. Based on the phase mechanism coupling state representation and mechanism partition state sequence, the sensor confidence gating value is calculated in the sensor confidence gating unit and input into the prediction reconstruction separation unit to generate the equipment state anomaly score and the sensor confidence anomaly score. Based on equipment status anomaly scoring, mechanism chain status representation, mechanism partition status sequence, sensor credibility anomaly scoring, and sensor credibility gating value, a stacker-reclaimer status monitoring information is constructed.

[0009] Optionally, the generation of the stacker-reclaimer state sequence includes: Intelligent sensors are installed on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism of the stacker-reclaimer to collect data such as slewing angle, vibration, temperature, structural strain, pitching angle, hydraulic pressure, traveling speed, bucket wheel current, and belt load. The slewing angle data, pitch angle data, travel speed data, bucket wheel current data, and belt load data are included in the basic motion data, and the vibration data, temperature data, hydraulic pressure data, and structural strain data are included in the equipment response data. The basic motion data and the equipment response data together form the operating status data. The operation status data is sampled according to a unified sampling period to obtain the unified sampled operation status data. Based on the sampling time corresponding to the unified sampling period, the running status data after unified sampling is time-aligned and written into the same multivariate time series record. Missing samples in the multivariate time series records are filled with missing samples, and duplicate samples in the multivariate time series records are merged to obtain the merged multivariate time series records. The merged multivariate time series records are subjected to dimensional unification processing to generate a stacker-reclaimer state sequence.

[0010] Optionally, constructing the phase chain MTAD-GAT model includes: The feature map attention structure, temporal map attention structure, prediction structure, and reconstruction structure in the MTAD-GAT model are invoked. Based on the feature map attention structure, the attention calculation method oriented towards all variables is transformed into a partitioned attention calculation method oriented towards the rotary mechanism, pitch mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism, and a mechanism chain partitioned attention unit is constructed. Based on the time-map attention structure, a new working condition phase slicing unit is added, forming a connection form in which the stacker-reclaimer state sequence first enters the working condition phase slicing unit and then enters the phase transition coupling unit; Based on the time-map attention structure, the attention calculation method for continuous sampling moments is transformed into a processing method for intra-phase time change and inter-phase transition for different phase segments, and a phase transition coupling unit is constructed. A sensor confidence gating unit is added before the output processing paths of the predicted structure and the reconstructed structure; Based on the prediction structure and the reconstruction structure, a prediction-reconstruction separation unit is constructed, forming a connection form in which the sensor confidence gating value participates in the separation of the output of the prediction structure and the reconstruction structure. A phase chain MTAD-GAT model is constructed based on the mechanism chain partition attention unit, the working condition phase slicing unit, the phase transition coupling unit, the sensor confidence gating unit, and the prediction reconstruction separation unit.

[0011] Optionally, the partitioning mechanism's partition state sequence includes: Based on the stacker-reclaimer state sequence, the correspondence between the operating status data and each mechanism is determined according to the data collection sources of the operating status data on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism; Based on the correspondence between the operating status data and each mechanism, the rotation angle data, vibration data, temperature data and structural strain data collected on the rotary mechanism will be classified into the data sequence corresponding to the rotary mechanism to generate the rotary mechanism status sequence. The pitch angle data, hydraulic pressure data, vibration data, and structural strain data collected on the pitch mechanism will be incorporated into the pitch mechanism state sequence. The walking speed data, vibration data, and temperature data collected on the walking mechanism will be included in the walking mechanism state sequence; The bucket wheel current data, vibration data, and temperature data collected on the bucket wheel mechanism will be incorporated into the bucket wheel mechanism state sequence; The belt load data, vibration data, and temperature data collected on the belt conveyor mechanism will be incorporated into the belt conveyor mechanism state sequence; Based on the sampling time, the state sequences of the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism are synchronously arranged to generate the mechanism partition state sequence.

[0012] Optionally, the generation mechanism chain state representation includes: Based on the mechanism partition state sequence, the mechanism chain partition attention unit determines the mechanism transmission relationship according to the mechanism connection sequence during the operation of the stacker-reclaimer, including the material reclaiming load transmission relationship, attitude load transmission relationship, attitude coupling transmission relationship and whole machine displacement transmission relationship; Based on the mechanism transmission relationship, it is transformed into material load transmission edge, attitude load transmission edge, attitude coupling transmission edge and whole machine displacement transmission edge, and mechanism transmission edge is generated. Based on the mechanism transmission edge, a transmission type identifier is added to each mechanism transmission edge according to the power transmission direction, load transmission direction and vibration propagation direction. Based on the institutional partition state sequence, local attention relationships are calculated for the operating state data corresponding to the same institution in the institutional chain partition attention unit; Based on the mechanism transit edge and transit type identifier, the mechanism chain partition attention unit calculates cross-mechanism attention relationship for the running state data between mechanism partition state sequences connected by mechanism transit edges. Based on local attention relationships and cross-institutional attention relationships, the operational state data corresponding to the same institution are weighted and summarized, and the operational state data with institutional transit edges are weighted and summarized to generate an institutional chain state representation.

[0013] Optionally, the generation of the intra-phase state representation includes: Based on the basic motion data in the stacker-reclaimer state sequence, the changes in slewing angle, pitch angle, travel speed, bucket wheel current, and belt load are determined in the working condition phase slice unit according to the data difference between adjacent sampling times. Based on the variation range of basic motion data in the stable section of the normal operation sample of the stacker-reclaimer, the low variation range of slewing angle, low variation range of pitch angle, low variation range of travel speed, low load variation range of bucket wheel current, and low load variation range of belt load are determined. When the change in walking speed is not within the low change range of walking speed, and the change in bucket wheel current is within the low load change range of bucket wheel current and the change in belt load are within the low load change range of belt load, the working condition phase slicing unit will determine the corresponding sampling time as the no-load moving phase. When the bucket wheel current change continues to rise during continuous sampling time, and the belt load change enters the belt load change range from the low load change range, the operation start phase is determined. When the changes in bucket wheel current and belt load remain within the belt load variation during continuous sampling time, and the changes in slewing angle are within the operating range, a stable material feeding phase is determined. When the pitch angle change is not within the low pitch angle change range and the bucket wheel current change does not form a continuous load change, determine the attitude adjustment phase; When the changes in slewing angle, pitch angle, travel speed, bucket wheel current, and belt load fall within the low range of slewing angle, pitch angle, travel speed, bucket wheel current, and belt load, the shutdown holding phase is determined. Based on the idle movement phase, operation entry phase, stable material picking phase, attitude adjustment phase, and shutdown holding phase corresponding to each sampling time, the stacker-reclaimer state sequence is divided to generate idle movement phase segments, operation entry phase segments, stable material picking phase segments, attitude adjustment phase segments, and shutdown holding phase segments. The changes in operating state data between adjacent sampling times within each phase segment are calculated to generate a state representation within the phase.

[0014] Optionally, the generation of the phase mechanism coupling state representation includes: Based on the state representation within the phase, the phase transition coupling unit determines the adjacent phase connection relationships between the no-load moving phase segment, the operation cutting phase segment, the stable material picking phase segment, the attitude adjustment phase segment, and the stop holding phase segment. Based on the adjacent phase connection relationship, it is transformed into a cutting transition edge, a load-stable transition edge, an attitude correction transition edge, and a shutdown transition edge, thus generating a phase transition edge; Based on the phase transition edge, the changes in the operating state data of adjacent working condition phase segments in the phase state representation are matched to determine the phase transition relationship corresponding to the phase transition edge. Based on the phase transition relationship, the attention weights corresponding to the mechanism transmission relationship in the mechanism chain state representation are adjusted in the phase transition coupling unit. Based on the adjusted mechanism chain state representation, the phase transition relationship corresponding to the phase transition edge is coupled with the running state data of the corresponding mechanism transmission edge in the mechanism chain state representation to generate a phase mechanism coupled state representation.

[0015] Optionally, the generation of device status anomaly scores and sensor reliability anomaly scores includes: Based on the phase mechanism coupled state representation and the mechanism partition state sequence, the prediction and reconstruction separation unit generates state prediction results and state reconstruction results according to the phase mechanism coupled state representation. The state prediction results correspond to the running state data at the next sampling time in the mechanism partition state sequence, and the state reconstruction results correspond to the running state data within the current continuous sampling window in the mechanism partition state sequence. Based on the state prediction results and the institutional partition state sequence, field-by-field difference calculation is performed to generate the prediction result deviation; Based on the state reconstruction results and the institutional partition state sequence, field-by-field difference calculation is performed to generate the reconstruction result deviation; Based on the institutional partition state sequence, the sensor confidence gating unit calculates the sensor confidence gating value according to the synchronicity of changes in basic action data and equipment response data, response lag time and consistency of change direction within the same institutional partition state sequence. Based on the prediction result deviation, reconstruction result deviation, sensor confidence threshold, and mechanism transmission relationship, the prediction and reconstruction separation unit separates the prediction result deviation and reconstruction result deviation to generate equipment status anomaly score and sensor confidence anomaly score.

[0016] Optionally, the status monitoring information for the stacker-reclaimer includes: Input the normal operation sample of the stacker-reclaimer into the phase chain MTAD-GAT model to generate the equipment status anomaly score and sensor reliability anomaly score corresponding to the normal operation sample of the stacker-reclaimer, and determine the equipment status anomaly judgment value and sensor reliability anomaly judgment value. Based on the equipment status anomaly score, mechanism chain status representation, and mechanism partition status sequence, the equipment status anomaly scores are summarized to obtain the scores for the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism. When the equipment status abnormality score exceeds the equipment status abnormality judgment value, the mechanism zone corresponding to the highest score among the slewing mechanism score, pitching mechanism score, traveling mechanism score, bucket wheel mechanism score, and belt conveyor mechanism score will be determined as the abnormal mechanism zone. Based on the sensor reliability anomaly score, sensor reliability gate value, and mechanism partition state sequence, the sensor reliability anomaly scores are summarized to obtain the sensor anomaly score corresponding to each smart sensor. When the sensor confidence anomaly score exceeds the sensor confidence anomaly judgment value, the anomaly location is sorted based on the sensor anomaly score and sensor confidence threshold value corresponding to each smart sensor to determine the location of the abnormal sensor. Based on the abnormal mechanism partition and the equipment status abnormality score, the abnormal equipment status results are generated by associating them according to the same sampling time. Based on the location of abnormal sensors and the abnormal sensor credibility score, the results are associated according to the same sampling time and combined to generate abnormal sensor credibility results. Based on abnormal mechanism partitions, abnormal sensor locations, abnormal equipment status results, and abnormal sensor reliability results, a stacker-reclaimer status monitoring information is constructed.

[0017] The beneficial effects of this invention are: First, this invention processes the state sequence of the mechanism partition through the mechanism chain partition attention unit, constructs the mechanism transmission relationship according to the power transmission direction, load transmission direction and vibration propagation direction of the stacker-reclaimer, and generates the mechanism chain state representation based on the local attention relationship and cross-mechanism attention relationship. It can incorporate the state transmission relationship between the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism into the anomaly judgment process, thereby improving the positioning accuracy of the abnormal mechanism partition and reducing misjudgments caused by relying solely on single sensor data or ordinary multi-source data fusion.

[0018] Secondly, this invention determines the working condition phase corresponding to each sampling moment in the stacker-reclaimer state sequence through the working condition phase slicing unit, and divides the stacker-reclaimer state sequence into no-load movement phase segment, operation entry phase segment, stable material reclaiming phase segment, attitude adjustment phase segment, and stop holding phase segment. Then, the phase transition relationship is coupled with the mechanism chain state representation through the phase transition coupling unit, which can distinguish between normal working condition phase changes and abnormal equipment changes, reduce false alarms caused by sudden state changes during operation entry, attitude adjustment, and stop holding, and improve the adaptability of the state monitoring results to complex operating conditions.

[0019] Furthermore, this invention calculates the sensor reliability gating value through a sensor reliability gating unit, and separates the prediction result deviation and reconstruction result deviation based on the sensor reliability gating value through a prediction reconstruction separation unit, generating equipment status anomaly scores and sensor reliability anomaly scores respectively. This enables the differentiation between equipment status anomalies and sensor reliability anomalies, thereby determining the abnormal mechanism zones and abnormal sensor locations, reducing the interference of sensor drift, loose installation, and sampling anomalies on equipment status judgment, and improving the reliability of stacker-reclaimer status monitoring information. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of a stacker-reclaimer status monitoring method based on intelligent sensors proposed in this invention; Figure 2 This is a schematic diagram of the phase chain MTAD-GAT model in this invention; Figure 3 This is a schematic diagram of the mechanism transmission relationship and the phase coupling processing of the working condition in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 A method for monitoring the status of a stacker-reclaimer based on intelligent sensors includes the following steps: Based on the intelligent sensors installed on the stacker-reclaimer, the operating status data is collected. The operating status data includes basic motion data and equipment response data. The basic motion data includes rotation angle data, pitch angle data, travel speed data, bucket wheel current data, and belt load data. The equipment response data includes vibration data, temperature data, hydraulic pressure data, and structural strain data. Data preprocessing is performed on the operating status data to generate the stacker-reclaimer status sequence. The data preprocessing includes time alignment, missing sample completion, duplicate sample merging, and dimension unification processing. Based on the feature map attention structure, temporal map attention structure, prediction structure and reconstruction structure of the MTAD-GAT model, a phase chain MTAD-GAT model is constructed. The phase chain MTAD-GAT model consists of mechanism chain partition attention unit, working condition phase slice unit, phase transition coupling unit, sensor confidence gating unit and prediction reconstruction separation unit. Based on the stacker-reclaimer state sequence, according to the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism corresponding to the operating state data, the stacker-reclaimer state sequence is divided into slewing mechanism state sequence, pitching mechanism state sequence, traveling mechanism state sequence, bucket wheel mechanism state sequence, and belt conveyor mechanism state sequence, generating mechanism partition state sequences. Based on the mechanism partition state sequence, in the mechanism chain partition attention unit, the mechanism transmission relationship is constructed according to the power transmission direction, load transmission direction and vibration propagation direction of the stacker-reclaimer. Local attention relationship is calculated for the running state data within the same mechanism partition state sequence. Then, based on the mechanism transmission relationship, cross-mechanism attention relationship is calculated for the running state data between adjacent mechanism partition state sequences, generating a mechanism chain state representation containing mechanism transmission relationship. Based on the basic motion data in the stacker-reclaimer state sequence, the working condition phase slicing unit is called to determine the working condition phase corresponding to each sampling moment in the stacker-reclaimer state sequence according to the changes in rotation angle, pitch angle, travel speed, bucket wheel current, and belt load. The stacker-reclaimer state sequence is then divided into no-load movement phase segment, operation entry phase segment, stable material reclaiming phase segment, attitude adjustment phase segment, and stop holding phase segment according to the working condition phase. The time change relationship within each working condition phase segment is calculated to generate the state representation within the phase. Based on the phase state representation, in the phase transition coupling unit, the phase transition relationship between the no-load moving phase segment, the operation cutting phase segment, the stable material picking phase segment, the attitude adjustment phase segment and the stop holding phase segment is calculated, and the phase transition relationship is coupled with the mechanism chain state representation to generate the phase mechanism coupling state representation. Based on the phase mechanism coupling state representation and mechanism partition state sequence, in the sensor confidence gating unit, the sensor confidence gating value is calculated according to the cooperative change relationship between the equipment response data and the basic action data within the same mechanism partition state sequence. The prediction reconstruction separation unit separates the prediction result deviation and reconstruction result deviation according to the sensor confidence gating value, and generates equipment state anomaly score and sensor confidence anomaly score. Based on equipment status anomaly scores, mechanism chain status representations, and mechanism partition status sequences, abnormal mechanism partitions are determined. Abnormal sensor locations are determined based on sensor reliability anomaly scores and sensor reliability threshold values. Equipment status anomaly results and sensor reliability anomaly results are generated based on equipment status anomaly scores and sensor reliability anomaly results. Stacker-reclaimer status monitoring information containing abnormal mechanism partitions, abnormal sensor locations, equipment status anomaly results, and sensor reliability anomaly results is constructed.

[0023] In this embodiment, the generation of the stacker-reclaimer state sequence includes: Intelligent sensors are installed on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism of the stacker-reclaimer. The slewing mechanism collects slewing angle data, vibration data, temperature data, and structural strain data; the pitching mechanism collects pitch angle data, hydraulic pressure data, vibration data, and structural strain data; the traveling mechanism collects traveling speed data, vibration data, and temperature data; the bucket wheel mechanism collects bucket wheel current data, vibration data, and temperature data; and the belt conveyor mechanism collects belt load data, vibration data, and temperature data. The slewing angle data, pitch angle data, travel speed data, bucket wheel current data, and belt load data are included in the basic motion data, and the vibration data, temperature data, hydraulic pressure data, and structural strain data are included in the equipment response data. The basic motion data and the equipment response data together form the operating status data. The operation status data is sampled according to a unified sampling period, which is determined by the lowest sampling frequency in the operation status data. The sampled data with a sampling period higher than the unified sampling period is downsampled according to the sampling time to obtain the unified sampled operation status data. Based on the sampling time corresponding to the unified sampling period, the running status data after unified sampling is time-aligned, and the basic action data and equipment response data under the same sampling time are written into the same multivariate timing record. Missing samples in multivariate time series records are imputed. When a missing sample is located between adjacent valid sampling times, it is imputed by interpolation using the previous adjacent valid sample value and the next adjacent valid sample value. When a missing sample is located at the start sampling time or the end sampling time, it is imputed by adjacent valid sample values. Repeated sampling merging is performed on repeated samples in multivariate time series records. Multiple sampled values ​​at the same sampling time and under the same data field are averaged to obtain merged multivariate time series records. The merged multivariate time series records are processed to unify the dimensions. The historical operating range is determined according to the historical minimum and maximum values ​​of the same data field. The corresponding data fields are converted into dimensionless data to generate the stacker-reclaimer state sequence.

[0024] In this embodiment, constructing the phase chain MTAD-GAT model includes: The feature map attention structure, temporal map attention structure, prediction structure, and reconstruction structure in the MTAD-GAT model are used as the foundation for building the phase chain MTAD-GAT model. Based on the feature map attention structure, the attention calculation method oriented towards all variables in the feature map attention structure is transformed into a partitioned attention calculation method oriented towards the rotary mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism. A mechanism chain partitioned attention unit is constructed, and the internal connection path of the same mechanism and the transmission connection path of adjacent mechanisms are set in the mechanism chain partitioned attention unit. Based on the time-map attention structure, a working condition phase slice unit is added to the phase chain MTAD-GAT model. The output of the working condition phase slice unit is connected to the input of the phase transition coupling unit, forming a connection form in which the stacker-reclaimer state sequence first enters the working condition phase slice unit and then enters the phase transition coupling unit. Based on the time-map attention structure, the attention calculation method oriented towards continuous sampling time in the time-map attention structure is transformed into a phase-in time change processing method and phase-to-phase transition processing method oriented towards idle movement phase segment, operation entry phase segment, stable material picking phase segment, attitude adjustment phase segment and stop holding phase segment, and a phase transition coupling unit is constructed. Connect the output of the mechanism chain partition attention unit to the input of the phase transition coupling unit to construct a connection path in which the mechanism chain state representation participates in the processing of the phase transition coupling unit; Before the output processing paths of the prediction structure and the reconstruction structure, a sensor confidence gating unit is added. The output of the phase transition coupling unit is connected to the input of the sensor confidence gating unit, and the output of the sensor confidence gating unit is connected to the input of the prediction reconstruction separation unit. Based on the prediction structure and the reconstruction structure, a prediction-reconstruction separation unit is constructed. The output of the prediction structure and the output of the reconstruction structure are connected to the input of the prediction-reconstruction separation unit. The output of the sensor confidence gating unit is connected to the input of the prediction-reconstruction separation unit, forming a connection form in which the sensor confidence gating value participates in the separation of the output of the prediction structure and the reconstruction structure. A phase chain MTAD-GAT model is constructed based on the mechanism chain partitioning attention unit, the working condition phase slicing unit, the phase transition coupling unit, the sensor credibility gating unit, and the prediction reconstruction separation unit. During the training of the phase chain MTAD-GAT model, historical operating status data, historical maintenance records, historical sensor calibration records, and historical anomaly handling records of the stacker-reclaimer are collected and merged according to the same stacker-reclaimer number, the same sampling time, and the same operating period to form a historical training sample set. Historical operating status data includes basic motion data and equipment response data. Historical maintenance records are used to mark periods of abnormal equipment status, and historical sensor calibration records are used to mark periods of abnormal sensor reliability. Data preprocessing consistent with the stacker-reclaimer state sequence is performed on the historical training sample set to generate a historical stacker-reclaimer state sequence. Furthermore, historical mechanism partition state sequences are generated according to the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism. The historical stacker-reclaimer state sequence is divided into continuous sampling windows, each containing 60 consecutive sampling moments. The first 59 sampling moments are used as training input, the 60th sampling moment is used as the state prediction target, and the complete continuous sampling window is used as the state reconstruction target. The phase chain MTAD-GAT model generates state prediction results and state reconstruction results based on the training input. The mean squared deviation between the state prediction result and the state prediction target is used as the prediction loss, and the mean squared deviation between the state reconstruction result and the state reconstruction target is used as the reconstruction loss. The sensor reliability threshold is generated based on the coordinated change relationship between basic motion data and equipment response data within the historical institutional partition state sequence. When the basic motion data changes continuously, and the equipment response data changes continuously in the same direction within a set response interval, the corresponding sensor reliability is marked as normal reliability. When the equipment response data changes abruptly, and the basic motion data within the same institutional partition state sequence does not show a corresponding change, and there are corresponding abnormal records in the historical sensor verification records, the corresponding sensor reliability is marked as abnormal reliability. The set response interval is determined by the average interval between the time when the basic motion data changes and the time when the equipment response data follows suit in the historical normal operation state data. The total loss of the phase chain MTAD-GAT model consists of prediction loss, reconstruction loss, mechanism transit consistency loss, phase transition consistency loss, and sensor confidence separation loss. Prediction loss is generated by the mean squared deviation between the state prediction result and the state prediction target; reconstruction loss is generated by the mean squared deviation between the state reconstruction result and the state reconstruction target; mechanism transit consistency loss is generated based on the difference between the attention weights of adjacent mechanism transit connection paths and non-adjacent mechanism connection paths in the mechanism chain partition attention unit, increasing the loss when the attention weight of a non-adjacent mechanism connection path is higher than that of an adjacent mechanism transit connection path; phase transition consistency loss is generated based on the difference between the transition weights of adjacent phase segments and non-adjacent phase segments in the phase transition coupling unit, increasing the loss when the transition weight of a non-adjacent phase segment is greater than that of an adjacent phase segment; sensor confidence separation loss is generated based on the cross-entropy deviation between the sensor confidence gating value and the normal confidence marker and the abnormal confidence marker. During training, the prediction loss, reconstruction loss, mechanism transmission consistency loss, phase transition consistency loss, and sensor confidence separation loss are summed according to fixed weights to obtain the total loss. The prediction loss weight is set to 1.0, the reconstruction loss weight is set to 1.0, the mechanism transmission consistency loss weight is set to 0.3, the phase transition consistency loss weight is set to 0.3, and the sensor confidence separation loss weight is set to 0.5. The Adam optimizer is used to iteratively update the parameters of the phase chain MTAD-GAT model. The initial learning rate is set to 0.001, the batch size is set to 32, and the maximum number of training epochs is set to 100. After each training epoch, the validation loss is calculated using the validation sample set. Training is stopped when the decrease in validation loss is less than 0.0001 for 10 consecutive training epochs. The model parameters corresponding to the minimum validation loss are saved to obtain the trained phase chain MTAD-GAT model.

[0025] In this embodiment, the partitioning mechanism's partition state sequence includes: Based on the stacker-reclaimer state sequence, the correspondence between the operating status data and each mechanism is determined according to the data collection sources of the operating status data on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism; Based on the correspondence between the operating status data and each mechanism, the rotation angle data, vibration data, temperature data and structural strain data collected on the rotary mechanism will be assigned to the data sequence corresponding to the rotary mechanism according to the sampling time, and a rotary mechanism status sequence will be generated. Based on the correspondence between the operational status data and each mechanism, the pitch angle data, hydraulic pressure data, vibration data and structural strain data collected on the pitch mechanism are assigned to the data sequence corresponding to the pitch mechanism according to the sampling time, and a pitch mechanism status sequence is generated. Based on the correspondence between the operational status data and each mechanism, the walking speed data, vibration data and temperature data collected on the walking mechanism are assigned to the data sequence corresponding to the walking mechanism according to the sampling time, and a walking mechanism status sequence is generated. Based on the correspondence between the operational status data and each mechanism, the bucket wheel current data, vibration data and temperature data collected on the bucket wheel mechanism are assigned to the data sequence corresponding to the bucket wheel mechanism according to the sampling time, and a bucket wheel mechanism status sequence is generated. Based on the correspondence between the operating status data and each mechanism, the belt load data, vibration data and temperature data collected on the belt conveyor mechanism will be assigned to the data sequence corresponding to the belt conveyor mechanism according to the sampling time, and a belt conveyor mechanism status sequence will be generated. Based on the sampling time, the state sequences of the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism are synchronously arranged to generate the mechanism partition state sequence.

[0026] In this embodiment, the generation mechanism chain state representation includes: Based on the mechanism partition state sequence, the mechanism chain partition attention unit determines the mechanism transmission relationship according to the mechanism connection sequence during the operation of the stacker-reclaimer. The mechanism transmission relationship includes the material reclaiming load transmission relationship between the bucket wheel mechanism and the belt conveyor mechanism, the attitude load transmission relationship between the pitch mechanism and the bucket wheel mechanism, the attitude coupling transmission relationship between the slewing mechanism and the pitch mechanism, and the overall displacement transmission relationship between the traveling mechanism and the slewing mechanism. Based on the mechanism transmission relationship, the material loading transmission relationship is transformed into a material loading transmission edge connecting the state sequence of the bucket wheel mechanism and the state sequence of the belt conveyor mechanism; the attitude load transmission relationship is transformed into an attitude load transmission edge connecting the state sequence of the pitch mechanism and the state sequence of the bucket wheel mechanism; the attitude coupling transmission relationship is transformed into an attitude coupling transmission edge connecting the state sequence of the slewing mechanism and the state sequence of the pitching mechanism; and the whole machine displacement transmission relationship is transformed into a whole machine displacement transmission edge connecting the state sequence of the traveling mechanism and the state sequence of the slewing mechanism, thus generating the mechanism transmission edge. Based on the mechanism transmission edge, a transmission type identifier is added to each mechanism transmission edge according to the power transmission direction, load transmission direction and vibration propagation direction. The transmission type identifier includes power transmission identifier, load transmission identifier and vibration propagation identifier. Based on the mechanism partition state sequence, in the mechanism chain partition attention unit, local attention relationships are calculated for the operating state data corresponding to the same mechanism. The local attention relationships are limited to the state sequence of the slewing mechanism, the state sequence of the pitching mechanism, the state sequence of the traveling mechanism, the state sequence of the bucket wheel mechanism, and the state sequence of the belt conveyor mechanism. Specifically, the operating state data corresponding to the same sampling time in the same mechanism state sequence are converted into attention calculation vectors, the similarity between the attention calculation vectors corresponding to the operating state data is calculated, the similarity is normalized to obtain the local attention weight, and the operating state data in the same mechanism state sequence are weighted and summed according to the local attention weight to obtain the local attention relationship. Based on the mechanism transfer edge and transfer type identifier, the mechanism chain partition attention unit calculates the cross-mechanism attention relationship for the running state data between mechanism partition state sequences connected by mechanism transfer edges. The cross-mechanism attention relationship is limited to the relationship between the bucket wheel mechanism state sequence and the belt conveyor mechanism state sequence, the pitch mechanism state sequence and the bucket wheel mechanism state sequence, the slewing mechanism state sequence and the pitch mechanism state sequence, and the traveling mechanism state sequence and the slewing mechanism state sequence. Specifically, the running state data corresponding to the same sampling time in the two mechanism partition state sequences connected by mechanism transfer edges are converted into attention calculation vectors, and the transfer type identifier is converted into an identifier vector with the same dimension as the attention calculation vector. The attention calculation vector and the identifier vector corresponding to the two mechanism partition state sequences are used together for similarity calculation. The similarity is normalized to obtain the cross-mechanism attention weight, and the running state data in the two mechanism partition state sequences connected by mechanism transfer edges are weighted and summed according to the cross-mechanism attention weight to obtain the cross-mechanism attention relationship. Based on local attention relationships and cross-institutional attention relationships, the institutional chain partitioning attention unit uses local attention relationships as the weighting basis for the operational state data corresponding to the same institution, and performs weighted summarization on the operational state data corresponding to the same institution. It also uses cross-institutional attention relationships as the weighting basis for the operational state data connected by institutional transmission edges, and performs weighted summarization on the operational state data connected by institutional transmission edges. Finally, it concatenates the weighted summarization results at the same sampling time with the identifier vector corresponding to the transmission type identifier to generate an institutional chain state representation that includes institutional transmission relationships.

[0027] In this embodiment, the generation of the phase-in-state representation includes: Based on the basic motion data in the stacker-reclaimer state sequence, the slewing angle data, pitch angle data, travel speed data, bucket wheel current data, and belt load data at the same sampling moment are extracted in the working condition phase slice unit. The changes in slewing angle, pitch angle, travel speed, bucket wheel current, and belt load are determined according to the data difference between adjacent sampling moments. Based on the variation range of basic motion data in the stable section of the stacker-reclaimer's normal operation sample, the low variation range of slewing angle, pitch angle, travel speed, bucket wheel current, and belt load is determined. The stacker-reclaimer's normal operation sample refers to the operation status data selected from the stacker-reclaimer's historical operation status data that does not show any abnormal equipment status, sensor reliability abnormalities, or corresponding historical maintenance records. The stacker-reclaimer's normal operation sample comes from the historical operation status data formed before the stacker-reclaimer was put into status monitoring and has been verified by manual maintenance records, historical abnormality handling records, and sensor calibration records. When the change in walking speed is not within the low change range of walking speed, and the change in bucket wheel current is within the low load change range of bucket wheel current and the change in belt load are within the low load change range of belt load, the working condition phase slicing unit will determine the corresponding sampling time as the no-load moving phase. When the bucket wheel current change continues to rise during the continuous sampling time, and the belt load change enters the belt load change range from the low load change range, the working condition phase slicing unit will determine the corresponding continuous sampling time as the operation entry phase. The belt load change range is determined by the change range of belt load data in the material reclaiming operation section in the normal operation sample of the stacker-reclaimer. When the changes in bucket wheel current and belt load remain within the continuous sampling time, and the changes in slewing angle are within the operating range, the working condition phase slicing unit will determine the corresponding continuous sampling time as the stable material taking phase. The operating range is determined by the range of change of slewing angle data in the material taking operation section in the normal operation sample of the stacker-reclaimer. When the pitch angle change is not within the low pitch angle change range and the bucket wheel current change does not form a continuous load change, the working condition phase slice unit will determine the corresponding sampling time as the attitude adjustment phase. When the slewing angle change falls within the low range of slewing angle change, the pitch angle change falls within the low range of pitch angle change, the travel speed change falls within the low range of travel speed change, the bucket wheel current change falls within the low load range of bucket wheel current change, and the belt load change falls within the low load range of belt load, the working condition phase slicing unit will determine the corresponding sampling time as the stop holding phase. Based on the idle movement phase, operation entry phase, stable material picking phase, attitude adjustment phase and stop holding phase corresponding to each sampling time, the stacker-reclaimer state sequence is divided according to the continuous identical working condition phases, generating idle movement phase segments, operation entry phase segments, stable material picking phase segments, attitude adjustment phase segments and stop holding phase segments. Based on the no-load movement phase segment, the operation entry phase segment, the stable material picking phase segment, the attitude adjustment phase segment, and the stop holding phase segment, the changes in the operating status data at adjacent sampling times within each working condition phase segment are calculated, and the changes in the operating status data within each working condition phase segment are arranged in the order of sampling time to generate the state representation within the phase.

[0028] In this embodiment, the generation phase mechanism coupling state representation includes: Based on the phase state representation, the phase transition coupling unit determines the adjacent phase connection relationship between the no-load moving phase segment, the operation cutting phase segment, the stable material reclaiming phase segment, the attitude adjustment phase segment, and the shutdown holding phase segment according to the order of each working condition phase segment in the stacker-reclaimer state sequence. Based on the adjacent phase connection relationship, the connection relationship from the unloaded moving phase segment to the operation cutting phase segment is transformed into a cutting transition edge, the connection relationship from the operation cutting phase segment to the stable material picking phase segment is transformed into a load-bearing stable transition edge, and the connection relationship from the stable material picking phase segment to the attitude adjustment phase segment is transformed into an attitude correction transition edge, and the connection relationship from the attitude adjustment phase segment to the stop holding phase segment is transformed into a stop transition edge, thus generating phase transition edges. Based on the phase transition edge, the changes in the operating state data of adjacent working condition phase segments in the phase state representation are matched to determine the phase transition relationship corresponding to the cut-in transition edge, the load stabilization transition edge, the attitude correction transition edge and the shutdown transition edge. Based on the phase transition relationship, the attention weights corresponding to the mechanism transmission relationship in the mechanism chain state representation are adjusted in the phase transition coupling unit. Specifically, in the phase transition relationships corresponding to the operation entry phase segment and the stable material picking phase segment, an attention weight is added to the material picking load transfer side between the state sequence of the bucket wheel mechanism and the state sequence of the belt conveyor mechanism; in the phase transition relationships corresponding to the attitude adjustment phase segment, an attention weight is added to the attitude load transfer side between the state sequence of the pitch mechanism and the state sequence of the bucket wheel mechanism; in the phase transition relationships corresponding to the no-load movement phase segment, an attention weight is added to the overall machine displacement transfer side between the state sequence of the traveling mechanism and the state sequence of the slewing mechanism. Based on the adjusted mechanism chain state representation, the phase transition relationships corresponding to the cut-in transition edge, the load-stabilized transition edge, the attitude correction transition edge, and the shutdown transition edge are coupled with the running state data of the corresponding mechanism transmission edge in the mechanism chain state representation to generate a phase mechanism coupled state representation.

[0029] In this embodiment, the generation of device status anomaly scores and sensor reliability anomaly scores includes: Based on the phase mechanism coupled state representation and the mechanism partition state sequence, the prediction and reconstruction separation unit generates state prediction results and state reconstruction results according to the phase mechanism coupled state representation. The state prediction results correspond to the operating state data at the next sampling time in the mechanism partition state sequence, and the state reconstruction results correspond to the operating state data within the current continuous sampling window in the mechanism partition state sequence. Specifically, generating the state prediction results and state reconstruction results includes: inputting the phase mechanism coupled state representation into the prediction structure, and having the prediction structure output the various operating state data at the next sampling time according to the state change order within the current continuous sampling window to generate the state prediction results; and inputting the phase mechanism coupled state representation into the reconstruction structure, and having the reconstruction structure output the various operating state data at the corresponding sampling time according to the state arrangement order within the current continuous sampling window to generate the state reconstruction results. Based on the state prediction results and the institutional partition state sequence, the state prediction results are compared with the operating state data at the corresponding sampling time in the institutional partition state sequence, and the field-by-field difference is calculated to generate the prediction result deviation. Based on the state reconstruction results and the institutional partition state sequence, the field-by-field difference calculation is performed between the state reconstruction results and the operating state data in the corresponding continuous sampling window in the institutional partition state sequence to generate the reconstruction result deviation. Based on the institutional partition state sequence, the sensor reliability gating unit calculates the sensor reliability gating value according to the synchronicity of changes in basic action data and equipment response data within the same institutional partition state sequence, the response lag time, and the consistency of change direction. Synchronicity of change refers to the corresponding change in equipment response data within the allowed response time after a change in basic action data. The response lag time is determined by the interval between the sampling time when the basic action data changes and the sampling time when the corresponding change occurs in the equipment response data. Consistency of change direction means that when the value of the basic action data increases, the equipment response data increases synchronously, and when the value of the basic action data decreases, the equipment response data decreases synchronously. When calculating the sensor reliability gating value, equipment response data that meets the synchronicity of change is recorded as synchronously valid, equipment response data with a response lag time within the allowed response time is recorded as lag-valid, and equipment response data that meets the consistency of change direction is recorded as directional valid. The sensor reliability gating value is then determined according to the number of times synchronous, lag-valid, and directional valid conditions are met. The more conditions met, the higher the sensor reliability gating value; the fewer conditions met, the lower the sensor reliability gating value. Based on the normal operation samples of the stacker-reclaimer, the average interval between changes in the equipment response data after changes in the basic motion data within the same mechanism partition state sequence is statistically analyzed to determine the allowable response time. When the basic motion data changes and the equipment response data shows a corresponding change in the same direction within the allowable response time, the sensor confidence gating unit increases the sensor confidence gating value of the corresponding equipment response data. When a single equipment response data undergoes a sudden change and the basic motion data within the same mechanism partition state sequence does not show a corresponding change within the allowable response time, the sensor confidence gating unit decreases the sensor confidence gating value of the corresponding equipment response data. Based on the prediction result deviation, reconstruction result deviation, sensor confidence threshold, and mechanism transmission relationship, the prediction and reconstruction separation unit separates the prediction result deviation and reconstruction result deviation. Following the same sampling time and the same data field, the prediction result deviation and reconstruction result deviation are added together to obtain the total deviation of the corresponding operating state data. When the response data of multiple devices change synchronously along the mechanism transmission relationship and the corresponding sensor confidence threshold does not decrease, an abnormal device state score is generated based on the total deviation of the corresponding operating state data. When the response data of a single device shows a total deviation, the corresponding sensor confidence threshold decreases, and other operating state data within the same mechanism partition state sequence do not show coordinated changes, a sensor confidence anomaly score is generated based on the total deviation of the corresponding operating state data.

[0030] In this embodiment, the status monitoring information for the stacker-reclaimer includes: Input the normal operation samples of the stacker-reclaimer into the trained phase chain MTAD-GAT model, and generate equipment state anomaly scores and sensor reliability anomaly scores corresponding to the normal operation samples of the stacker-reclaimer according to the same processing procedure as the state sequence of the stacker-reclaimer. Based on the equipment state anomaly scores and sensor reliability anomaly scores corresponding to the normal operation samples of the stacker-reclaimer, determine the equipment state anomaly judgment value and the sensor reliability anomaly judgment value. The equipment state anomaly judgment value is determined by the upper boundary of the equipment state anomaly score distribution corresponding to the normal operation samples of the stacker-reclaimer, and the sensor reliability anomaly judgment value is determined by the upper boundary of the sensor reliability anomaly score distribution corresponding to the normal operation samples of the stacker-reclaimer. Based on equipment status anomaly scoring, mechanism chain status representation, and mechanism partition status sequence, the equipment status anomaly scores are summarized separately according to the slewing mechanism status sequence, pitching mechanism status sequence, traveling mechanism status sequence, bucket wheel mechanism status sequence, and belt conveyor mechanism status sequence to obtain slewing mechanism scores, pitching mechanism scores, traveling mechanism scores, bucket wheel mechanism scores, and belt conveyor mechanism scores. Specifically, the total deviations corresponding to the operating status data belonging to the slewing mechanism status sequence are averaged to obtain the slewing mechanism score; the total deviations corresponding to the operating status data belonging to the pitching mechanism status sequence are averaged to obtain the pitching mechanism score; the total deviations corresponding to the operating status data belonging to the traveling mechanism status sequence are averaged to obtain the traveling mechanism score; the total deviations corresponding to the operating status data belonging to the bucket wheel mechanism status sequence are averaged to obtain the bucket wheel mechanism score; and the total deviations corresponding to the operating status data belonging to the belt conveyor mechanism status sequence are averaged to obtain the belt conveyor mechanism score. When the equipment status abnormality score exceeds the equipment status abnormality judgment value, the mechanism zone corresponding to the highest score among the slewing mechanism score, pitching mechanism score, traveling mechanism score, bucket wheel mechanism score, and belt conveyor mechanism score will be determined as the abnormal mechanism zone. Based on sensor reliability anomaly scores, sensor reliability threshold values, and mechanism partition state sequences, the sensor reliability anomaly scores are summarized according to the smart sensors corresponding to rotation angle data, pitch angle data, travel speed data, bucket wheel current data, belt load data, vibration data, temperature data, hydraulic pressure data, and structural strain data, respectively, to obtain the sensor anomaly scores corresponding to each smart sensor. When the sensor confidence anomaly score exceeds the sensor confidence anomaly judgment value, based on the sensor anomaly score and sensor confidence threshold value corresponding to each smart sensor, the smart sensors are anomaly location sorted. Among them, the smart sensors with higher sensor anomaly scores and lower sensor confidence threshold values ​​are sorted higher. The setting position of the smart sensor ranked first on the stacker-reclaimer is determined as the abnormal sensor position. Based on the abnormal mechanism partition and the equipment status abnormality score, the abnormal equipment status results are generated by associating them according to the same sampling time. Based on the location of abnormal sensors and the abnormal sensor credibility score, the results are associated according to the same sampling time and combined to generate abnormal sensor credibility results. Based on abnormal mechanism partitions, abnormal sensor locations, abnormal equipment status results, and abnormal sensor reliability results, a stacker-reclaimer status monitoring information is constructed.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to the condition monitoring scenario of a stacker-reclaimer in a bulk material yard. This stacker-reclaimer is responsible for stacking and reclaiming bulk coal and ore for a long time. During operation, it involves the continuous coordination of the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism. The original condition monitoring method on site mainly relies on vibration threshold, temperature threshold, current threshold, and belt load threshold for alarms. When the bucket wheel cuts into the stockpile, the belt load suddenly increases, the pitch angle is adjusted, or the machine's traveling operation is converted to slewing operation, the data from multiple sensors will fluctuate simultaneously in a short period of time. The traditional method easily judges normal operating condition changes as abnormalities. When a vibration sensor is affected by dust covering or loose installation and causes a sudden change, the traditional method easily judges the sensor problem as equipment failure, resulting in unclear maintenance direction.

[0032] In this embodiment, intelligent sensors are deployed on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism of the stacker-reclaimer to collect data on slewing angle, pitching angle, traveling speed, bucket wheel current, belt load, vibration, temperature, hydraulic pressure, and structural strain. The collected operating status data is processed through time alignment, missing sample completion, duplicate sample merging, and dimensionless standardization to generate a stacker-reclaimer state sequence. This state sequence is divided into slewing mechanism state sequences, pitching mechanism state sequences, traveling mechanism state sequences, bucket wheel mechanism state sequences, and belt conveyor mechanism state sequences according to their mechanism origin, forming mechanism-specific state sequences. This allows subsequent monitoring to distinguish between data changes within the same mechanism and changes transmitted between different mechanisms.

[0033] In application, the mechanism chain partitioned attention unit in the phase chain MTAD-GAT model constructs mechanism transmission relationships based on the mechanism partition state sequence. It incorporates the material handling load transmission between the bucket wheel mechanism and the belt conveyor mechanism, the attitude load transmission between the pitch mechanism and the bucket wheel mechanism, the attitude coupling transmission between the slewing mechanism and the pitch mechanism, and the overall displacement transmission between the traveling mechanism and the slewing mechanism into the attention calculation scope. For vibration, temperature, hydraulic pressure, structural strain, and corresponding basic motion data within the same mechanism, local attention relationships are calculated; for mechanisms with mechanism transmission relationships, cross-mechanism attention relationships are calculated. The resulting mechanism chain state representation reflects the propagation path of anomalies between mechanisms, rather than simply recording whether a single sensor value exceeds a threshold.

[0034] The working condition phase slicing unit determines the working condition phase based on changes in rotation angle, pitch angle, travel speed, bucket wheel current, and belt load, dividing the stacker-reclaimer state sequence into idle movement phase segments, operation entry phase segments, stable reclaiming phase segments, attitude adjustment phase segments, and stop-and-hold phase segments. The phase transition coupling unit calculates the phase transition relationships between different working condition phase segments and couples these relationships with the mechanism chain state representation. For example, in the operation entry and stable reclaiming phase segments, the reclaiming load transfer weight between the bucket wheel mechanism and the belt conveyor mechanism is increased; in the attitude adjustment phase segment, the attitude load transfer weight between the pitch mechanism and the bucket wheel mechanism is increased. Through this processing, the model can distinguish between state fluctuations caused by normal working condition switching and continuous deviations caused by equipment anomalies.

[0035] The sensor reliability gating unit calculates the sensor reliability gating value based on the synchronicity of changes in basic action data and equipment response data within the same mechanism partition state sequence, as well as the consistency of response lag time and direction of change. When the bucket wheel current increases, and the bucket wheel mechanism vibration and belt load change in the same direction within the allowable response time, this change is identified as an equipment operation response. When a single vibration data point changes abruptly, but the bucket wheel current, belt load, and other equipment response data within the same mechanism do not show corresponding changes, this change is more likely to be identified as a sensor reliability anomaly. The prediction, reconstruction, and separation unit generates equipment state anomaly scores and sensor reliability anomaly scores based on the phase mechanism coupling state representation and mechanism partition state sequence, ultimately constructing stacker-reclaimer state monitoring information, including abnormal mechanism partitions, abnormal sensor locations, equipment state anomaly results, and sensor reliability anomaly results.

[0036] This embodiment selects 2400 sets of continuously operating samples, including 1800 sets of normal operation samples, 360 sets of equipment anomaly samples, and 240 sets of sensor anomaly samples. The equipment anomaly samples cover abnormal loads in the bucket wheel mechanism, abnormal pitch hydraulic pressure, abnormal vibrations in the slewing mechanism, and abnormal loads in the belt conveyor mechanism; the sensor anomaly samples cover loose vibration sensors, drifting temperature sensors, and abrupt changes in strain data. Training samples account for 70%, validation samples for 15%, and test samples for 15%. Comparison method one is the traditional fixed threshold monitoring method, which alarms based on fixed judgment values ​​for vibration, temperature, current, hydraulic pressure, and belt load; comparison method two is the conventional multivariate statistical anomaly detection method, which calculates anomaly scores after standardizing multi-source sensor data; method three is the method of this invention, namely a stacker-reclaimer status monitoring method based on intelligent sensors.

[0037] Table 1. Experimental Comparison Results of Stacker-Reclaimer Condition Monitoring Methods

[0038] As shown in Table 1, the equipment anomaly identification accuracy of the method of this invention is 90.8%, which is 12.2 percentage points higher than the traditional fixed threshold monitoring method and 6.5 percentage points higher than the conventional multivariate statistical anomaly detection method. This result is directly related to the processing method of the mechanism chain partition attention unit. The traditional fixed threshold monitoring method only judges whether a single data point has exceeded the limit, and cannot determine whether there is a material loading load transmission relationship between the increase in bucket wheel current, the increase in belt load, and the change in bucket wheel vibration. Although the conventional multivariate statistical anomaly detection method can integrate multiple data fields, it does not establish mechanism transmission relationships according to the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism. The method of this invention combines local attention relationships and cross-mechanism attention relationships, and can identify whether the response data of multiple devices change synchronously along the mechanism transmission relationship, thus making the judgment of real equipment anomalies more stable.

[0039] Regarding the accuracy of abnormal mechanism partition location, the method of this invention achieves 88.5%, which is higher than the 71.4% of the traditional fixed threshold monitoring method and the 79.2% of the conventional multivariate statistical anomaly detection method. In the test samples, the traditional fixed threshold monitoring method often judges the increase in belt load caused by bucket wheel mechanism abnormality as an abnormality of both the bucket wheel mechanism and the belt conveyor mechanism, resulting in an expanded location range; the conventional multivariate statistical anomaly detection method can detect anomalies in multiple fields, but it is difficult to determine the origin of the anomaly. The method of this invention is based on the mechanism chain state representation, incorporating the material loading load transfer edge, attitude load transfer edge, attitude coupling transfer edge, and whole machine displacement transfer edge into the state expression. When the anomaly spreads along the mechanism transmission relationship, it can retain the difference between the anomaly source and the transmission response, thereby improving the accuracy of abnormal mechanism partition location.

[0040] Regarding the accuracy of sensor reliability anomaly identification, the method of this invention achieves 86.2%, significantly higher than the 64.8% of the traditional fixed threshold monitoring method. This difference mainly stems from the processing of the sensor reliability gating unit. Traditional fixed threshold monitoring methods typically trigger an alarm directly as equipment malfunction when vibration data suddenly increases, failing to determine whether this vibration change is coordinated with changes in bucket wheel current, belt load, hydraulic pressure, and structural strain. Conventional multivariate statistical anomaly detection methods can identify abrupt changes in data, but still easily confuse sensor anomalies with equipment malfunctions. The method of this invention uses a sensor reliability gating value to assess the reliability of equipment response data. When a single piece of equipment response data shows a sudden change and other operating state data within the same institutional partition state sequence do not exhibit coordinated changes, a sensor reliability anomaly score can be generated, reducing the likelihood of misjudging sensor malfunctions as equipment malfunctions.

[0041] Regarding the false alarm rate during operating condition switching, the method of this invention has a rate of 6.7%, lower than the 16.9% of the traditional fixed threshold monitoring method and the 11.8% of the conventional multivariate statistical anomaly detection method. During the operation entry, stable material handling, and attitude adjustment processes of the stacker-reclaimer, the bucket wheel current, belt load, pitch angle, and hydraulic pressure frequently undergo continuous changes. Fixed threshold methods are sensitive to these changes and easily identify normal operation processes as abnormalities. The method of this invention divides the stacker-reclaimer state sequence into different operating condition phase segments through an operating condition phase slicing unit and calculates the phase transition relationship through a phase transition coupling unit. This allows the model to understand changes in operating state data according to different operating condition phases, thus significantly reducing false alarms caused by operating condition switching.

[0042] Regarding the average anomaly confirmation time, the method of this invention achieves 27.6 seconds, shorter than the 42.5 seconds of the traditional fixed threshold monitoring method and the 35.8 seconds of the conventional multivariate statistical anomaly detection method. This improvement is not achieved by simply shortening the sampling period, but because the phase mechanism coupling state representation, when generating the equipment status anomaly score and sensor confidence anomaly score, already incorporates the mechanism transmission relationship and the phase change of the operating condition simultaneously. For continuously expanding equipment anomalies, the response data of multiple devices will change synchronously along the mechanism transmission relationship, allowing the equipment status anomaly score to form a stable result earlier; for single sensor drift, the sensor confidence threshold value will decrease, enabling the sensor confidence anomaly score to be output more quickly.

[0043] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring the status of a stacker-reclaimer based on intelligent sensors, characterized in that, Includes the following steps: Based on the intelligent sensors installed on the stacker-reclaimer, the operating status data is collected, data preprocessing is performed, and a stacker-reclaimer status sequence is generated; The phase chain MTAD-GAT model is constructed, including mechanism chain partitioning attention unit, working condition phase slicing unit, phase transition coupling unit, sensor confidence gating unit and prediction reconstruction separation unit; Based on the content of each mechanism corresponding to the operating status data, the stacker-reclaimer status sequence is divided into mechanism partition status sequences; Based on the mechanism partition state sequence, mechanism transfer relations are constructed in the mechanism chain partition attention unit, local attention relations and cross-mechanism attention relations are calculated, and mechanism chain state representation is generated; Based on the stacker-reclaimer state sequence, the working condition phase slicing unit is called to determine the working condition phase, and the stacker-reclaimer state sequence is divided into different working condition phase segments to generate a state representation within the phase; Based on the phase state representation, the phase transition relationship between different working condition phase segments is calculated in the phase transition coupling unit to generate the phase mechanism coupling state representation. Based on the phase mechanism coupling state representation and mechanism partition state sequence, the sensor confidence gating value is calculated in the sensor confidence gating unit and input into the prediction reconstruction separation unit to generate the equipment state anomaly score and the sensor confidence anomaly score. Based on equipment status anomaly scoring, mechanism chain status representation, mechanism partition status sequence, sensor credibility anomaly scoring, and sensor credibility gating value, a stacker-reclaimer status monitoring information is constructed.

2. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The generated stacker-reclaimer state sequence includes: Intelligent sensors are installed on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism of the stacker-reclaimer to collect data such as slewing angle, vibration, temperature, structural strain, pitching angle, hydraulic pressure, traveling speed, bucket wheel current, and belt load. The slewing angle data, pitch angle data, travel speed data, bucket wheel current data, and belt load data are included in the basic motion data, and the vibration data, temperature data, hydraulic pressure data, and structural strain data are included in the equipment response data. The basic motion data and the equipment response data together form the operating status data. The operation status data is sampled according to a unified sampling period to obtain the unified sampled operation status data. Based on the sampling time corresponding to the unified sampling period, the running status data after unified sampling is time-aligned and written into the same multivariate time series record. Missing samples in the multivariate time series records are filled with missing samples, and duplicate samples in the multivariate time series records are merged to obtain the merged multivariate time series records. The merged multivariate time series records are subjected to dimensional unification processing to generate a stacker-reclaimer state sequence.

3. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The construction of the phase chain MTAD-GAT model includes: The feature map attention structure, temporal map attention structure, prediction structure, and reconstruction structure in the MTAD-GAT model are invoked. Based on the feature map attention structure, the attention calculation method oriented towards all variables is transformed into a partitioned attention calculation method oriented towards the rotary mechanism, pitch mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism, and a mechanism chain partitioned attention unit is constructed. Based on the time-map attention structure, a new working condition phase slicing unit is added, forming a connection form in which the stacker-reclaimer state sequence first enters the working condition phase slicing unit and then enters the phase transition coupling unit; Based on the time-map attention structure, the attention calculation method for continuous sampling moments is transformed into a processing method for intra-phase time change and inter-phase transition for different phase segments, and a phase transition coupling unit is constructed. A sensor confidence gating unit is added before the output processing paths of the predicted structure and the reconstructed structure; Based on the prediction structure and the reconstruction structure, a prediction-reconstruction separation unit is constructed, forming a connection form in which the sensor confidence gating value participates in the separation of the output of the prediction structure and the reconstruction structure. A phase chain MTAD-GAT model is constructed based on the mechanism chain partition attention unit, the working condition phase slicing unit, the phase transition coupling unit, the sensor confidence gating unit, and the prediction reconstruction separation unit.

4. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The partitioning mechanism's partitioning state sequence includes: Based on the stacker-reclaimer state sequence, the correspondence between the operating status data and each mechanism is determined according to the data collection sources of the operating status data on the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism and belt conveyor mechanism; Based on the correspondence between the operating status data and each mechanism, the rotation angle data, vibration data, temperature data and structural strain data collected on the rotary mechanism will be classified into the data sequence corresponding to the rotary mechanism to generate the rotary mechanism status sequence. The pitch angle data, hydraulic pressure data, vibration data, and structural strain data collected on the pitch mechanism will be incorporated into the pitch mechanism state sequence. The walking speed data, vibration data, and temperature data collected on the walking mechanism will be included in the walking mechanism state sequence; The bucket wheel current data, vibration data, and temperature data collected on the bucket wheel mechanism will be incorporated into the bucket wheel mechanism state sequence; The belt load data, vibration data, and temperature data collected on the belt conveyor mechanism will be incorporated into the belt conveyor mechanism state sequence; Based on the sampling time, the state sequences of the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism are synchronously arranged to generate the mechanism partition state sequence.

5. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The state representation of the generation mechanism chain includes: Based on the mechanism partition state sequence, the mechanism chain partition attention unit determines the mechanism transmission relationship according to the mechanism connection sequence during the operation of the stacker-reclaimer, including the material reclaiming load transmission relationship, attitude load transmission relationship, attitude coupling transmission relationship and whole machine displacement transmission relationship; Based on the mechanism transmission relationship, it is transformed into material load transmission edge, attitude load transmission edge, attitude coupling transmission edge and whole machine displacement transmission edge, and mechanism transmission edge is generated. Based on the mechanism transmission edge, a transmission type identifier is added to each mechanism transmission edge according to the power transmission direction, load transmission direction and vibration propagation direction. Based on the institutional partition state sequence, local attention relationships are calculated for the operating state data corresponding to the same institution in the institutional chain partition attention unit; Based on the mechanism transit edge and transit type identifier, the mechanism chain partition attention unit calculates cross-mechanism attention relationship for the running state data between mechanism partition state sequences connected by mechanism transit edges. Based on local attention relationships and cross-institutional attention relationships, the operational state data corresponding to the same institution are weighted and summarized, and the operational state data with institutional transit edges are weighted and summarized to generate an institutional chain state representation.

6. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The generated phase-in-state representation includes: Based on the basic motion data in the stacker-reclaimer state sequence, the changes in slewing angle, pitch angle, travel speed, bucket wheel current, and belt load are determined in the working condition phase slice unit according to the data difference between adjacent sampling times. Based on the variation range of basic motion data in the stable section of the normal operation sample of the stacker-reclaimer, the low variation range of slewing angle, low variation range of pitch angle, low variation range of travel speed, low load variation range of bucket wheel current, and low load variation range of belt load are determined. When the change in walking speed is not within the low change range of walking speed, and the change in bucket wheel current is within the low load change range of bucket wheel current and the change in belt load are within the low load change range of belt load, the working condition phase slicing unit will determine the corresponding sampling time as the no-load moving phase. When the bucket wheel current change continues to rise during continuous sampling time, and the belt load change enters the belt load change range from the low load change range, the operation start phase is determined. When the changes in bucket wheel current and belt load remain within the belt load variation during continuous sampling time, and the changes in slewing angle are within the operating range, a stable material feeding phase is determined. When the pitch angle change is not within the low pitch angle change range and the bucket wheel current change does not form a continuous load change, determine the attitude adjustment phase; When the changes in slewing angle, pitch angle, travel speed, bucket wheel current, and belt load fall within the low range of slewing angle, pitch angle, travel speed, bucket wheel current, and belt load, the shutdown holding phase is determined. Based on the idle movement phase, operation entry phase, stable material picking phase, attitude adjustment phase, and shutdown holding phase corresponding to each sampling time, the stacker-reclaimer state sequence is divided to generate idle movement phase segments, operation entry phase segments, stable material picking phase segments, attitude adjustment phase segments, and shutdown holding phase segments. The changes in operating state data between adjacent sampling times within each phase segment are calculated to generate a state representation within the phase.

7. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The coupling state representation of the generated phase mechanism includes: Based on the state representation within the phase, the phase transition coupling unit determines the adjacent phase connection relationships between the no-load moving phase segment, the operation cutting phase segment, the stable material picking phase segment, the attitude adjustment phase segment, and the stop holding phase segment. Based on the adjacent phase connection relationship, it is transformed into a cutting transition edge, a load-stable transition edge, an attitude correction transition edge, and a shutdown transition edge, thus generating a phase transition edge; Based on the phase transition edge, the changes in the operating state data of adjacent working condition phase segments in the phase state representation are matched to determine the phase transition relationship corresponding to the phase transition edge. Based on the phase transition relationship, the attention weights corresponding to the mechanism transmission relationship in the mechanism chain state representation are adjusted in the phase transition coupling unit. Based on the adjusted mechanism chain state representation, the phase transition relationship corresponding to the phase transition edge is coupled with the running state data of the corresponding mechanism transmission edge in the mechanism chain state representation to generate a phase mechanism coupled state representation.

8. The method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The generated device status anomaly score and sensor reliability anomaly score include: Based on the phase mechanism coupled state representation and the mechanism partition state sequence, the prediction and reconstruction separation unit generates state prediction results and state reconstruction results according to the phase mechanism coupled state representation. The state prediction results correspond to the running state data at the next sampling time in the mechanism partition state sequence, and the state reconstruction results correspond to the running state data within the current continuous sampling window in the mechanism partition state sequence. Based on the state prediction results and the institutional partition state sequence, field-by-field difference calculation is performed to generate the prediction result deviation; Based on the state reconstruction results and the institutional partition state sequence, field-by-field difference calculation is performed to generate the reconstruction result deviation; Based on the institutional partition state sequence, the sensor confidence gating unit calculates the sensor confidence gating value according to the synchronicity of changes in basic action data and equipment response data, response lag time and consistency of change direction within the same institutional partition state sequence. Based on the prediction result deviation, reconstruction result deviation, sensor confidence threshold, and mechanism transmission relationship, the prediction and reconstruction separation unit separates the prediction result deviation and reconstruction result deviation to generate equipment status anomaly score and sensor confidence anomaly score.

9. A method for monitoring the status of a stacker-reclaimer based on intelligent sensors according to claim 1, characterized in that, The status monitoring information for the stacker-reclaimer includes: Input the normal operation sample of the stacker-reclaimer into the phase chain MTAD-GAT model to generate the equipment status anomaly score and sensor reliability anomaly score corresponding to the normal operation sample of the stacker-reclaimer, and determine the equipment status anomaly judgment value and sensor reliability anomaly judgment value. Based on the equipment status anomaly score, mechanism chain status representation, and mechanism partition status sequence, the equipment status anomaly scores are summarized to obtain the scores for the slewing mechanism, pitching mechanism, traveling mechanism, bucket wheel mechanism, and belt conveyor mechanism. When the equipment status abnormality score exceeds the equipment status abnormality judgment value, the mechanism zone corresponding to the highest score among the slewing mechanism score, pitching mechanism score, traveling mechanism score, bucket wheel mechanism score, and belt conveyor mechanism score will be determined as the abnormal mechanism zone. Based on the sensor reliability anomaly score, sensor reliability gate value, and mechanism partition state sequence, the sensor reliability anomaly scores are summarized to obtain the sensor anomaly score corresponding to each smart sensor. When the sensor confidence anomaly score exceeds the sensor confidence anomaly judgment value, the anomaly location is sorted based on the sensor anomaly score and sensor confidence threshold value corresponding to each smart sensor to determine the location of the abnormal sensor. Based on the abnormal mechanism partition and the equipment status abnormality score, the abnormal equipment status results are generated by associating them according to the same sampling time. Based on the location of abnormal sensors and the abnormal sensor credibility score, the results are associated according to the same sampling time and combined to generate abnormal sensor credibility results. Based on abnormal mechanism partitions, abnormal sensor locations, abnormal equipment status results, and abnormal sensor reliability results, a stacker-reclaimer status monitoring information is constructed.

Images