Data-driven hair dryer motor fault prediction and operation monitoring method

Abstract

The present application relates to motor operation data processing and computer fault diagnosis technical field, specifically to the hair dryer motor fault prediction and operation monitoring method based on data driving, the method obtains control instruction data, driving feedback data and operation state data in operation period, constructs working condition description vector and response description vector, uses health operation sample to establish condition baseline model, generates the counterfactual health response corresponding to the current working condition, and forms the working condition decoupling residual sequence by actual response and counterfactual health response;According to the single cycle and the cross cycle two time scales, update the degradation memory, determine the motor health state and the candidate fault mode by combining the residual correlation path and the closed loop check.This application can separate the normal response and the sustained degradation response caused by gear switching, power supply fluctuation and stage change, keep the comparability of fault characteristics in different operation periods, inhibit the interference of accidental disturbance on degradation trajectory, and identify the motor fault precursor developed across the cycle.

Description

Technical Field

[0001] This invention relates to the field of motor operation data processing and computer fault diagnosis technology, specifically a data-driven method for predicting and monitoring the operation of a hair dryer motor. Background Technology

[0002] The operating status of a hair dryer motor is typically collected and recorded by the controller during motor startup, acceleration, speed stabilization, gear switching, deceleration, and shutdown. Common data includes speed commands, speed feedback, drive current, supply voltage, control duty cycle, commutation sequence, cumulative running time, start / stop status, and protection status. Existing monitoring schemes often employ fixed thresholds, segmented thresholds, or deviation judgment methods based on historical averages. These schemes use phenomena such as increased current, decreased speed, accelerated temperature rise, and increased duty cycle as anomaly indicators, and statistically analyze peak values, average values, fluctuation amplitudes, and rates of change based on single-run records or fixed time windows. When a statistic exceeds a preset range, the control program outputs an anomaly flag, or inputs multiple statistics into a pre-established classifier to determine whether the motor is in a normal, slightly abnormal, or faulty state. While this approach is relatively straightforward, its judgment criteria are often separate from specific operating conditions, making it difficult to reflect the normal response differences that the same motor should exhibit under different fan speeds, power supply conditions, continuous running times, and start / stop intervals.

[0003] To reduce the sensitivity of fixed thresholds to changes in operating conditions, existing data-driven solutions typically train time-series classification models or remaining life prediction models using healthy and faulty samples. Continuously acquired control and feedback quantities are segmented into samples of fixed length, and then the fault probability is obtained through feature extraction, sequence encoding, and category mapping. Some solutions also establish several sub-models based on different operating conditions, or perform normalization on the input data to reduce dimensional differences and sample amplitude variations. These solutions rely on the training samples' coverage of the actual operating condition distribution. When power supply fluctuations, short-term load changes, operating condition switching combinations, or incomplete operating phases occur in actual operation, which are less common in the samples, the model easily learns these operating condition differences as fault characteristics. Simultaneously, occasional disturbances in a single operation and degradation that continues to develop across multiple operating cycles may have similar local manifestations in the original data. Fixed time windows weaken the phase sequence and cross-cycle continuity, causing the prediction results to drift with changes in operating conditions.

[0004] The normal response fluctuations of a hair dryer motor under different control conditions are superimposed with the abnormal responses caused by the motor's own degradation. Existing monitoring methods based on uniform thresholds, fixed time windows, or original time-series samples lack a health reference corresponding to the current operating condition, thus failing to reliably separate operating condition factors from degradation factors. The same current increment, speed lag, or duty cycle compensation may be a normal control response under one operating condition, but may reflect a decline in motor performance under another. If judgment is made directly based on the original amplitude, normal gear switching, power supply changes, or short-term load disturbances are easily written into the fault trajectory. If the judgment threshold is raised, early degradation that accumulates slowly over multiple service cycles will be masked. The core difficulty caused by this is that the monitoring system cannot form a degradation characterization comparable to the operating conditions, nor can it confirm the true fault evolution based on the persistence direction and correlation sequence of anomalies across multiple operating cycles. Summary of the Invention

[0005] The purpose of this invention is to provide a data-driven method for predicting and monitoring the failure of a hair dryer motor, which can solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A data-driven method for predicting and monitoring hair dryer motor faults includes: acquiring motor control command data, drive feedback data, and operating status data for each operating cycle of the hair dryer; A working condition description vector is constructed based on the motor control command data and the operating status data, and a response description vector is constructed based on the drive feedback data. A conditional baseline model is established between the operating condition description vector and the corresponding health response description vector using healthy operating samples, and a counterfactual health response corresponding to the current operating condition is generated. A condition-decoupled residual sequence is formed based on the actual response and the counterfactual health response; The degradation memory is updated according to two time scales: single cycle and cross cycle. The motor health status and candidate fault modes are determined based on the matching relationship between the degradation memory and the pre-stored fault mode trajectory.

[0007] Preferably, the construction of the operating condition description vector includes: using start-up, acceleration, speed stabilization, gear shifting, deceleration and shutdown events as data segment boundaries within the operating cycle, and encoding the speed command changes, control duty cycle changes, power supply voltage fluctuations, cumulative running time and start-stop intervals in each segment into stage operating condition sub-vectors; The stage connection identifier is generated according to the sequence of each segment, and the stage condition sub-vector is combined with the stage connection identifier to form a condition description vector that maintains the order of the operation process.

[0008] Preferably, the conditional baseline model includes a working condition neighborhood retrieval layer and a response reconstruction layer, wherein the working condition neighborhood retrieval layer follows the control intention between the working condition description vectors. Figure 1 The healthy operating samples are screened based on consistency, operational phase consistency, and power supply status consistency. The response reconstruction layer reconstructs the counterfactual health response under the current operating condition based on the response description vector of the screened healthy operating samples, and retains the contribution identifier of each healthy operating sample to the counterfactual health response.

[0009] Preferably, the operating condition decoupling residual sequence includes speed following residual, drive current residual, control compensation residual, commutation timing residual and shutdown attenuation residual; A residual association graph is established according to the operating stage, generation order and contribution identifier of each residual. Residuals that appear consecutively within the same operating cycle are encoded as short-term memory states, and residuals that maintain the same direction and repeat the associated path across operating cycles are encoded as long-term degenerate memory states.

[0010] Preferably, the generation of the stage condition sub-vector includes: mapping similar operation segments with different durations to a unified stage process coordinate system; Extract the relative time shift, local slope inflection, and compensation amount pullback position between the control command and the drive feedback on the unified stage process coordinates; An incomplete identifier is set for the operating segment that is interrupted by gear switching or protection action, and the incomplete identifier and the interruption position are retained in the operating condition description vector, without replacing the interrupted part with supplementary data.

[0011] Preferably, the working condition neighborhood retrieval layer performs bidirectional consistency verification on the healthy operation samples. First, it retrieves candidate healthy operation samples from the current working condition description vector, and then retrieves historical working condition description vectors adjacent to the current operating cycle from each candidate healthy operation sample. Only candidate healthy running samples that satisfy the consistency of the running stage arrangement in both forward and reverse retrieval are input into the response reconstruction layer, and the remaining candidate healthy running samples are recorded as boundary samples.

[0012] Preferably, the coverage confidence of the current counterfactual health response is calculated based on the contribution identifier, and the coverage confidence is combined with the path integrity of the residual association graph to form a memory update weight; When the healthy operation samples are concentrated in a single operating condition area or when there is a stage missing in the residual correlation path, the corresponding residuals are written into the isolated memory area. Only after the working condition neighborhood expands or the associated path is completed in a subsequent operating cycle will the residual in the isolated memory region be incorporated into the long-term degraded memory state.

[0013] Preferably, the long-term degenerate memory state is set with path memory slots according to the paths in the residual association graph, and each path memory slot records the occurrence period, directional consistency and stage migration order of the corresponding residual combination; When the same residual first appears in the startup phase and migrates to the acceleration or steady-state phase in subsequent operating cycles, the path memory slots before and after the migration are connected to form a fault evolution chain, and the candidate fault mode is updated according to the correspondence between different fault evolution chains and candidate fault modes.

[0014] Preferably, for each residual sequence written into the long-term degraded memory state, a playback anchor point is selected based on the neighborhood relationship between the current operating condition description vector and the historical healthy operation sample; Keeping the current actual response unchanged, the first residual sequence and the second residual sequence are formed by taking the counterfactual health response generated by the conditional baseline model and the health response corresponding to the replay anchor point as references, respectively. Based on the consistency between the first residual sequence and the second residual sequence in terms of residual direction, generation order, and cross-period continuity, it is determined whether the corresponding residual sequence is retained in the long-term degenerate memory state or regressed to the short-term memory state.

[0015] Preferably, cross-cycle closed-loop verification is performed on the residual sequence retained in the long-term degraded memory state, the first residual sequence and the second residual sequence are reverse-mapped to the corresponding historical operating cycle, and the operating condition description vector, counterfactual health response and memory update record in each historical operating cycle are read; For historical operating cycles that show residual direction reversal or sequence breakage after reverse mapping, the corresponding long-term degradation memory write is cancelled. For historical operating cycles that maintain continuity, the write is retained. Based on the verified long-term degradation memory state, a motor health state sequence and a candidate fault mode sequence are generated.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. A condition description vector is constructed based on motor control command data and operating status data, and a response description vector is constructed based on drive feedback data. A conditional baseline between the operating condition and the healthy response is established using healthy operating samples, generating a corresponding counterfactual healthy response for the current operating condition. The actual response is no longer directly compared with a unified threshold, but rather the difference is calculated with the healthy response under the same operating condition. The resulting speed following residual, drive current residual, control compensation residual, commutation timing residual, and shutdown attenuation residual have a unified operating condition reference. Single-cycle memory is used to retain abnormal combinations within the current operation, while cross-cycle memory is used to accumulate residuals with consistent directions and recurring associated paths. Occasional disturbances will not be directly converted into long-term degradation records due to a single deviation, and slow degradation will not be filtered out by a fixed threshold due to a small single deviation. The conditional baseline, operating condition decoupling residual, and dual-timescale memory work together to create distinguishable trajectories between normal operating condition changes and continuous motor degradation at the data level, solving the problem of incomparable fault representations under different operating conditions.

[0017] 2. Operational data is categorized into startup, acceleration, stabilization, gear shifting, deceleration, and shutdown events, and mapped to a unified stage process coordinate system. This allows for comparison of similar operational stages with varying durations based on relative progress, preserving relative time shifts, local slope transitions, and compensation pullback positions between stages. Healthy samples undergo bidirectional consistency verification via forward and reverse retrieval before participating in counterfactual response reconstruction. When the operating condition neighborhood coverage is insufficient or the residual path is incomplete, the corresponding data enters an isolated memory area, awaiting further expansion of the operating condition neighborhood or completion of the associated path before being incorporated into long-term degradation records. Residual correlation graphs, path memory slots, and fault evolution chains are used to preserve the order of anomaly generation and stage transition relationships. Replay anchor points and cross-cycle closed-loop verification are used to verify the direction and continuity of residuals and to undo historical writes involving direction reversals or sequence breaks. This processing maintains the operational stage sequence, sample coverage boundaries, and historical memory traceability, reducing interference from unknown operating conditions and incomplete operational records on the health state sequence and candidate fault mode sequence. Attached Figure Description

[0018] Figure 1 A flowchart for the overall process of predicting and monitoring the faults of a hair dryer motor; Figure 2 Flowcharts for phased chemical condition coding and counterfactual health response are constructed; Figure 3 This is a graph showing the evolution of runtime residuals under the stage process coordinates; Figure 4 A distribution diagram of healthy neighborhood distance and contribution weight; Figure 5 Flowchart for decoupling residuals, degradation memory, and closed-loop verification under operating conditions; Figure 6 The component diagram of the decoupled residual vector for the operating condition. Detailed Implementation

[0019] In one embodiment, a data-driven method for predicting and monitoring the faults of a hair dryer motor is provided, comprising: acquiring motor control command data, drive feedback data, and operating status data for each operating cycle of the hair dryer; constructing a condition description vector based on the motor control command data and operating status data, and constructing a response description vector based on the drive feedback data; establishing a conditional baseline model between the condition description vector and the corresponding health response description vector using healthy operating samples, and generating a counterfactual health response corresponding to the current operating condition; forming a condition decoupling residual sequence based on the actual response and the counterfactual health response; updating the degradation memory according to two time scales, single-cycle and cross-cycle, and determining the motor health status and candidate fault modes based on the matching relationship between the degradation memory and the pre-stored fault mode trajectory.

[0020] In this embodiment, the data from when the motor receives the start control to when the motor enters a stable shutdown state is defined as an operating cycle. The motor control command data includes speed command, gear switching command and drive control quantity. The drive feedback data includes speed feedback, drive current, control compensation quantity and commutation timing. The operating status data includes power supply voltage, cumulative running time, start and stop status, protection status and shutdown interval between adjacent operating cycles. All types of data are accompanied by a unified time stamp generated by the control program and are written into the cycle data record according to the operating cycle identifier.

[0021] The periodic data record adopts a storage format that links control command sequences, feedback sequences, state sequences, and data quality identifiers. The control command sequence is used to express the target operating state that the motor should execute at the current moment. The feedback sequence is used to express the actual response of the motor to the control command. The state sequence is used to express the power supply and operating background of the current response. The data quality identifier is used to record situations such as data missing, operation phase truncation, protection action intervention, and discontinuous time stamps. The missing data position retains a null value identifier and does not directly replace the original missing data with adjacent values.

[0022] Please refer to Figure 1 After the operation monitoring program reads a complete operating cycle or a completed partial operating phase, it first identifies the event type of the control process, and then encodes the control intent, operating status and event sequence into a condition description vector. Subsequently, the condition baseline model generates a response description vector that a healthy motor should form under the same operating conditions. The difference between the actual response and the counterfactual healthy response is converted into residual nodes. The residual nodes are connected according to their occurrence stage and sequence to form a residual association graph. The degradation memory is updated in the current cycle based on the continuity of the residual association graph, and continues to perform cross-cycle updates when the next operating cycle arrives.

[0023] In this embodiment, the start event is determined when the speed command changes from the stopped state to the non-stop state and the speed feedback begins to change continuously; the acceleration event is determined when the speed command remains non-decreasing while the speed feedback remains increasing; the speed stabilization event is determined when the speed command is in a stable range and the speed feedback changes around the corresponding target state; the gear shifting event is determined when the speed command undergoes a discrete change; the deceleration event is determined when the speed command decreases and the speed feedback decreases accordingly; and the stop event is determined when the speed command returns to the stopped state and continues until the speed feedback enters the stopped stable range.

[0024] In a preferred embodiment, please refer to Figure 2 The construction of the operating condition description vector includes: using start-up, acceleration, speed stabilization, gear shifting, deceleration, and shutdown events as data segment boundaries within the operating cycle; encoding the speed command changes, control duty cycle changes, power supply voltage fluctuations, cumulative running time, and start-stop intervals in each segment into stage operating condition sub-vectors; generating stage connection identifiers according to the sequential relationship of each segment; and combining the stage operating condition sub-vectors with the stage connection identifiers to form an operating condition description vector that maintains the order of the operating process.

[0025] When identifying data segment boundaries, the control program does not use changes in a single sampling point as the basis for event switching. Instead, it checks the direction of change in speed command, the direction of change in speed feedback, and the direction of change in control duty cycle within continuous data intervals. When the event conditions of adjacent data intervals change stably, the segment boundary is determined. For control commands that change back and forth near the boundary, the continuous changes are merged into a transition interval, and the transition interval is assigned to the corresponding operating stage after the change ends, so as to avoid the same gear adjustment being split into multiple events without independent physical meaning.

[0026] The stage operating condition sub-vector includes stage type encoding, stage duration state, speed command start and end state, speed command change pattern, control duty cycle reference state, control duty cycle change pattern, power supply voltage center state, power supply voltage fluctuation pattern, cumulative running time state, and start / stop interval state. The change pattern is not represented by a single maximum value, but is composed of the relative changes of the stage start point, stage middle, stage end point, and major turning points, so that stages with different durations can retain the internal evolution structure of the control process.

[0027] The stage connection identifier is used to record the type of the preceding stage, the type of the succeeding stage, whether there is a control command jump between stages, whether adjacent stages are separated by protection actions, and whether the current stage is the start or end stage of the operating cycle. The stage connection identifier and the stage operating condition sub-vector are written into the ordered operating condition record according to the order of event occurrence, so that the operating process of stopping directly after startup, entering steady speed after startup, and continuously shifting gears after startup forms different operating condition descriptions.

[0028] To ensure that similar operation phases with different durations enter a unified comparison space, this embodiment converts the original time position within a phase into phase process coordinates, which are calculated according to the following formula: ; in, This indicates the relative process of the current data point within its respective runtime phase, where t represents the uniform time stamp of the current data point. Indicates the start time marker of the corresponding operational phase. The sign indicates the end time of the current running phase, and the division sign indicates the ratio between the running time of the current phase and the total duration of the phase.

[0029] When the start time of a certain acceleration phase is marked as 20 and the end time is marked as 60, the data point marked as 20 corresponds to... The data point with a value of 0 and a time stamp of 40 corresponds to The data point with a value of 0.5 and a time stamp of 60 corresponds to... The value is 1. Even if the actual duration of another acceleration phase is different, it is mapped according to the process range of 0 to 1. This allows for comparison of speed following state, control compensation state and drive current state at the same relative process position, instead of directly taking the difference in phase duration as the difference in motor degradation.

[0030] Please refer to Figure 3 , Figure 3 The horizontal axis uses Chinese segmentation to represent the phased process, with the initial segment, early rising segment, middle rising segment, late rising segment, near steady state segment, and final segment sequentially representing the process position within the same operating phase. The vertical axis represents the residual intensity of each monitoring channel under a unified process coordinate system. Figure 3 Corresponding process coordinates The constructed results show that the curves represent the speed following residual, drive current residual, control compensation residual, commutation timing residual, and shutdown attenuation residual, respectively. Figure 3 The two horizontal reference lines in the diagram represent the first and second warning thresholds, respectively. Because... By mapping acceleration, steady-state, or shutdown phases of different durations to the same comparison space, each point on the curve can be regarded as the residual response of the same phase at the same process position, avoiding the phase length deviation caused by comparing only according to the original time sequence.

[0031] from Figure 3As can be seen, the speed following residual increases slowly from the initial stage to the middle stage, accelerates significantly after entering the near-steady-state stage, and exceeds the first warning threshold near the end stage, indicating that the motor speed closed loop has shown a sustainable deviation in the later stage. The drive current residual is generally higher than the control compensation residual, and increases in the same direction as the speed following residual after the later stage, indicating that load, winding heating, or friction changes have a cumulative effect on current demand. The commutation timing residual shows fine oscillations after the middle stage, but does not exceed the second warning threshold for a long time, indicating that this channel is more suitable as an early fluctuation indication rather than a sole basis for fault diagnosis.

[0032] Furthermore, the generation of the stage operating condition sub-vector includes: mapping similar operating segments with different durations to a unified stage process coordinate; extracting the relative time shift, local slope transition, and compensation amount withdrawal position between the control command and drive feedback on the unified stage process coordinate; setting an incomplete identifier for operating segments truncated by gear switching or protection action, and retaining the incomplete identifier and truncation position in the operating condition description vector, without replacing the truncated part with supplementary data.

[0033] The relative time shift is obtained by comparing the starting positions of the main change range of the speed command and the main change range of the speed feedback on the stage process coordinates. The local slope turning point is obtained by detecting the process position where the change direction or rate of change of the speed feedback changes stably. The compensation amount pullback position is obtained by detecting the process position where the control duty cycle reaches a local high position and then begins to fall continuously during acceleration. The relative time shift, local slope turning point and compensation amount pullback position are all expressed in stage process coordinates and are separate from the actual stage duration.

[0034] When a running segment is truncated due to gear shifting, the incomplete identifier records the reason for the truncation as a change in control commands, and saves the stage process coordinates at the time of the truncation. When a running segment is truncated due to a protection action, the incomplete identifier records the reason for the truncation as a change in protection status, and saves the control commands and feedback segments that were formed before the protection action. When the conditional baseline model searches for healthy running samples, it only considers samples with the same integrity status or that can cover the currently completed process interval as candidates, and does not regard the truncated unknown response as an observed response.

[0035] In this embodiment, the response description vector is generated according to the same stage boundary as the operating condition description vector. The response description vector of each stage includes the speed feedback trajectory, drive current trajectory, control compensation trajectory, commutation timing trajectory, and speed decay trajectory during the shutdown stage. The trajectory data is extracted at a fixed relative position on a unified stage process coordinate system, and the main turning points and stage end states are retained, so that the conditional baseline model can reconstruct the complete trajectory and also reconstruct the local trajectory that has been completed in the truncated stage.

[0036] In a preferred embodiment, the conditional baseline model includes a working condition neighborhood retrieval layer and a response reconstruction layer, wherein the working condition neighborhood retrieval layer follows the control intention between working condition description vectors. Figure 1 The healthy operating samples are screened based on consistency, operational phase consistency, and power supply status consistency. The response reconstruction layer reconstructs the counterfactual healthy response under the current operating condition based on the response description vector of the screened healthy operating samples, and retains the contribution identifier of each healthy operating sample to the counterfactual healthy response.

[0037] The healthy operation sample consists of historical operating cycles within the health status confirmation interval. The health status confirmation interval is determined by the period in which there is no protection status in the continuous operation record, the residual correlation path does not form cross-cycle accumulation, and the playback verification results are consistent. Each healthy operation sample saves the operating condition description vector, response description vector, stage integrity identifier, and sample source cycle. The condition baseline model does not directly mix the raw data of multiple operating cycles, but keeps the operating condition and response of each sample corresponding.

[0038] The operating condition neighborhood retrieval layer calculates the operating condition distance between the current operating condition description vector and the j-th healthy operating sample, and the operating condition distance is calculated according to the following formula: ; in, Let represent the distance of the j-th healthy operating sample relative to the current operating condition, and m represent the number of operating condition features included in the comparison. This represents the q-th feature in the current operating condition description vector. This represents the corresponding feature in the j-th healthy operating sample. This represents the comparison weight of the q-th working condition feature. This represents the discrete scale of the corresponding feature in the healthy operating sample, and the absolute value sign indicates the non-negative amplitude of the feature difference. Indicates the current stage sequence Sequence of sample stages The order difference between them, b represents the constraint weight of the order difference in the working condition distance.

[0039] When the current operating cycle has the same startup, acceleration, and stabilization sequence as a certain healthy operating sample... If the healthy operating sample lacks a gear shifting phase that has already occurred in the current operating cycle, then... Take a value greater than 0. When the speed command state and the power supply voltage state are similar but the stage sequence is different, the sequence difference term still makes the corresponding value greater than 0. As the value increases, the working condition neighborhood retrieval layer excludes healthy operating samples that are only similar in numerical amplitude but have different control processes.

[0040] Control Figure 1Consistency is determined by comparing stage type, speed command change direction, gear shift position, and stop command position. Operation stage consistency is determined by comparing stage arrangement, stage integrity, and stage connection identifier. Power supply status consistency is determined by comparing the center state and fluctuation pattern of power supply voltage. Only healthy operating samples that meet the retrieval conditions in all three types of consistency are included in the candidate sample set. The candidate sample set is arranged in ascending order of operating condition distance and the corresponding distance is retained.

[0041] Preferably, the working condition neighborhood retrieval layer performs bidirectional consistency verification on the healthy operation samples. First, candidate healthy operation samples are retrieved from the current working condition description vector. Then, for each candidate healthy operation sample, historical working condition description vectors adjacent to the current operating cycle are retrieved in reverse. Only candidate healthy operation samples that satisfy the consistency of the operating phase arrangement in both forward and reverse retrieval are input into the response reconstruction layer, and the remaining candidate healthy operation samples are recorded as boundary samples.

[0042] During reverse retrieval, the working condition description vector of the candidate healthy operating sample is used as the retrieval center. The system searches for the neighboring operating cycles in the historical working condition records before the current operating cycle. If the reverse retrieval result contains the working condition neighborhood of the current operating cycle and the stage arrangement obtained by the reverse retrieval is consistent with the current operating cycle, then the candidate healthy operating sample passes the bidirectional consistency check. If the candidate healthy operating sample is only close to the current working condition in the forward retrieval, but the reverse retrieval results in another type of stage arrangement, then it is determined to be an accidental close sample on the neighborhood boundary.

[0043] Boundary samples do not participate in the main reconstruction of the current counterfactual health response, but retain the condition distance, stage difference location and response description vector of the boundary samples. When the current condition neighborhood is expanded in subsequent running cycles and the same stage arrangement appears repeatedly, the bidirectional consistency check is re-executed on the boundary samples. If the boundary sample passes the re-check, it enters the candidate sample set. If it fails the re-check, it continues to maintain the boundary state and its historical records are not directly deleted.

[0044] The response reconstruction layer calculates the sample contribution weight based on the healthy operating samples that have passed the bidirectional consistency check, and reconstructs the counterfactual health response corresponding to the current operating condition according to the sample contribution weight. The calculation method is as follows:

[0045] ; in, Let represent the contribution weight of the j-th healthy operating sample, and exp represent the exponential operation with the natural constant as the base. This represents the set of candidate samples that have passed the bidirectional consistency check. This represents the operating condition distance of the k-th healthy operating sample within the set. This represents the counterfactual health response description vector corresponding to the current operating condition. Let represent the response description vector of the j-th healthy operating sample, and the summation symbol represents the positional combination of each healthy response according to its contribution weight.

[0046] When the candidate sample set contains two healthy operating samples with corresponding operating condition distances of 0.3 and 1.1, the sample with a distance of 0.3 receives a larger score. Samples with a distance of 1.1 still retain a small contribution. The counterfactual health response is formed by the combination of the speed feedback trajectory, drive current trajectory, control compensation trajectory and commutation timing trajectory of the two samples at the same stage of the process position. The contribution weight, source period and trajectory interval participating in the reconstruction of any healthy operating sample are written into the contribution identifier.

[0047] Please refer to Figure 4 , Figure 4 The horizontal axis represents the Chinese sorting position of the healthy neighborhood sample in the current working condition search window, the vertical axis represents the calculated neighborhood value, and the curves show the working condition distances. Contribution weight And the changes in cumulative contribution, with dotted markers indicating valid health samples selected for the counterfactual health response construction. Figure 4 Simultaneously corresponding working condition distance With contribution weight Two steps: the first measures the proximity of the current stage to historical health stages in the condition coding space; the second assigns reconstruction weights based on the distance. The horizontal axis does not use numerical indices but instead displays samples in relative proximity, making... Figure 4 The curve directly reflects the screening process of health references from strong correlation to weak correlation.

[0048] from Figure 4 It can be seen from the first few healthy samples Keep it in the lower range, corresponding The cumulative contribution curve is significantly higher than that of later samples, rising rapidly in the first half and gradually stabilizing, indicating that the counterfactual health response is mainly supported by a few highly similar health cycles. As the sample ranking moves further down the spectrum... Gradually increase The decline continues, and although later samples still participate in the normalization calculation, their actual impact on the reconstruction results is small. Effective labels are concentrated in positions with small distances and high weights, indicating that the system does not simply average historical data, but rather forms a traceable weighted reference based on the similarity of operating conditions.

[0049] For a complete running phase, the response reconstruction layer generates counterfactual health responses from the start point to the end point of the phase. For a truncated running phase, the response reconstruction layer only generates counterfactual health responses from the start point to the truncation position and saves the comparable interval in the counterfactual health response description vector. The actual response and the counterfactual health response only perform residual calculation within the comparable interval. The predicted health response value beyond the truncation position is not written into the current residual sequence.

[0050] In a preferred embodiment, please refer to Figure 5 The decoupling residual sequence of the operating conditions includes speed following residual, drive current residual, control compensation residual, commutation timing residual and shutdown attenuation residual; a residual association graph is established according to the operating stage, generation order and contribution identifier of each residual, and the residuals that appear continuously in the same operating cycle are encoded as short-term memory state, and the residuals that keep the direction consistent and the associated path repeats across operating cycles are encoded as long-term degradation memory state.

[0051] The speed following residual is formed by the difference between the actual speed feedback trajectory and the counterfactual healthy speed feedback trajectory on the same stage process coordinate. The drive current residual is formed by the difference between the actual drive current trajectory and the counterfactual healthy drive current trajectory. The control compensation residual is formed by the difference between the actual control compensation trajectory and the counterfactual healthy control compensation trajectory. The commutation timing residual is formed by the difference between the actual commutation interval sequence and the counterfactual healthy commutation interval sequence at the corresponding speed state. The shutdown attenuation residual is formed by the difference between the actual speed attenuation trajectory and the counterfactual healthy attenuation trajectory during the shutdown stage.

[0052] The decoupling residual description vector of the current operating condition at the c-th comparable position is generated according to the following formula: ; in, This represents the residual description vector at the c-th comparable position. , , , and These represent, in sequence, the actual speed feedback, drive current, control compensation amount, commutation timing status, and shutdown attenuation status. , , , and The matrix sequentially represents the actual speed feedback, drive current, control compensation, commutation timing state, and shutdown attenuation state in the counterfactual health response. The subtraction in the matrix represents the directional deviation of the actual state from the healthy state calculated under the same process position and the same operating condition reference.

[0053] When the actual speed feedback is lower than the counterfactual healthy speed feedback, the speed following component in the residual vector takes a negative value. When the actual drive current is higher than the counterfactual healthy drive current, the drive current component takes a positive value. If the speed following component is continuously negative and the drive current component is continuously positive in the same acceleration phase, the residual correlation diagram records the two as residual nodes with a stable directional combination in the same phase, instead of directly writing a single instantaneous deviation into long-term degradation memory.

[0054] Please refer to Figure 6 , Figure 6 The horizontal axis represents the Chinese segment position of the sampling process within a stage, and the vertical axis represents the residual component amplitude after decoupling of the operating conditions. The five curves correspond to... The components include the speed following component, drive current component, control compensation component, commutation timing component, and shutdown attenuation component. Figure 6 With residual vector Correspondingly, the displayed value is the fractional deviation obtained by subtracting the counterfactual health response from the current observed response, rather than the original sensor value. Therefore, the positive or negative direction of the curve represents whether the relative health reference is too high or too low, while the duration of the curve and the synchronization relationship are used to determine whether the anomaly has a stable mechanistic orientation.

[0055] from Figure 6 It is evident that the speed following component remains negative after the mid-rise stage, while the drive current component remains positive simultaneously. These two components form a deviation combination that is opposite in direction but synchronized in time, consistent with common response characteristics when the load increases, bearing friction rises, or motor efficiency decreases. The control compensation component exhibits a stepped increase in the later rise stage, indicating that the controller is increasing the compensation to offset insufficient output. The commutation timing component shows local oscillations, suggesting a slight disturbance in electromagnetic commutation or drive timing. The shutdown decay component only rises significantly in the final stage, indicating that the inertial decay anomaly is more of an auxiliary evidence in the later stages.

[0056] The residual correlation graph uses a combination of operating stage and residual type as node identifiers. Each node stores the residual direction, residual duration interval, residual trajectory shape, corresponding contribution identifier, and sample coverage status. Within the same stage, time edges are connected according to the order of residual generation. Adjacent stages are connected by migration edges according to stage connection identifiers. Rotation speed following residuals are connected with drive current residuals and control compensation residuals by correlation edges according to common process intervals. Commutation timing residuals are connected with other residual nodes according to their corresponding rotation speed status.

[0057] The short-term memory state stores residual paths that have been formed within the current running cycle but have not yet been confirmed across cycles. The short-term memory records the combination of residual nodes, the stage at which the nodes appear, the start and end positions of the path, and the combination of directions. When a certain residual appears only in a local position within the same running cycle and is restored to the counterfactual healthy response neighborhood in a subsequent stage, the short-term memory retains the local anomaly record but does not establish a cross-cycle path. When residuals in the same direction continue to appear in subsequent stages, the short-term memory connects adjacent residual nodes as a path to be confirmed.

[0058] The long-term degenerate memory state uses the residual paths that recur across operating cycles as the basic storage objects. When the path to be confirmed in the current operating cycle is consistent with the path in the historical short-term memory in terms of residual type, direction combination, stage order, and migration position, the path to be confirmed enters the long-term degenerate memory update process. If the paths are only similar in numerical magnitude but different in residual direction or stage order, they are stored in different path records and are not merged.

[0059] Furthermore, the coverage confidence of the current counterfactual health response is calculated based on the contribution identifier, and the coverage confidence is combined with the path integrity of the residual association graph as the memory update weight; when the healthy operation samples are concentrated in a single operating condition area or there is a stage missing in the residual association path, the corresponding residual is written into the isolated memory area; only after the operating condition neighborhood is expanded or the association path is completed in subsequent operating cycles, the residuals in the isolated memory area are merged into the long-term degradation memory state.

[0060] The coverage confidence of the counterfactual health response is calculated based on the contribution distribution of candidate healthy operating samples, the bidirectional consistency check results, and the residual path integrity. The coverage confidence is determined according to the following formula: ; in, This indicates the coverage confidence level in the nth running cycle. This represents the set of candidate samples corresponding to the nth running cycle. Represents the reciprocal of the working distance. This indicates the degree of completeness of the observed nodes relative to the expected nodes in the current residual correlation path. This indicates the bidirectional consistency check status. It is set to 1 when the check passes and to a value between 0 and 1 when there are boundary samples to be checked. The multiplication indicates that the sample coverage, path integrity and bidirectional consistency jointly constrain the current memory write.

[0061] When the contributions of candidate samples are concentrated in samples with small operating condition distances, the residual correlation path has covered the startup and acceleration phases, and the bidirectional consistency check passes, A higher value is achieved when all candidate samples are located at the boundary of the operating condition neighborhood or when the current running record is truncated during the acceleration phase. Cumulative items or decline, As a result, the degradation memory update program distinguishes between residual paths with complete healthy references and residual paths with insufficient reference coverage.

[0062] The isolation memory area stores residual description vectors, residual association paths, operating condition description vectors, candidate healthy operating samples, and isolation reasons according to the operating cycle. Isolation reasons include concentrated healthy samples, insufficient bidirectional consistency, missing operating phases, and broken residual paths. When the subsequent operating cycle arrives, the program re-searches for healthy operating samples adjacent to the isolated operating condition and checks whether the same residual node reappears in the complete phase. The isolation memory is only released when the operating condition neighborhood is expanded or the residual path is completed.

[0063] In a preferred embodiment, the long-term degradation memory state is set with path memory slots according to the paths in the residual association graph. Each path memory slot records the occurrence period, directional consistency and stage migration order of the corresponding residual combination. When the same residual first appears in the startup stage and migrates to the acceleration stage or steady-state stage in subsequent operating cycles, the path memory slots before and after the migration are connected to form a fault evolution chain, and the candidate fault mode is updated according to the correspondence between different fault evolution chains and candidate fault modes.

[0064] The path memory slot adopts a recording structure that links path identifiers, residual node sets, start stages, end stages, direction sequences, occurrence cycle sets, continuous occurrence states, operating condition neighborhood identifiers, and memory values. The path identifier is generated by the combination of residual types and stage order. Paths with the same residual type but different stage order use different path identifiers, and paths with the same stage order but different residual directions also use different path identifiers, thereby maintaining the directional information of the fault evolution path.

[0065] When a negative residual in speed following and a positive residual in drive current occur during the startup phase, and the steady-state phase of the same operating cycle recovers to the counterfactual healthy response neighborhood, the program records one occurrence in the startup phase path memory slot. When the same directional combination continues into the acceleration phase in subsequent operating cycles, the startup phase path memory slot and the acceleration phase path memory slot establish a migration connection. When the control compensation positive residual enters the steady-state phase in subsequent operating cycles, the program adds a steady-state phase path memory slot after the original migration connection, forming a fault evolution chain that extends from the startup phase to the acceleration phase and then to the steady-state phase.

[0066] The path memory slot's memory value in the nth runtime is updated according to the following formula: ; in, This represents the memory value of path p after the nth running cycle. This represents the memory value of path p after the previous execution cycle. This represents the retention coefficient determined based on the interval between adjacent occurrences and directional continuity. This represents the complete state of path p in the current residual correlation graph. This indicates the directional consistency state of path p in the current running cycle. Multiplication means that the current residual is only accumulated to the corresponding path memory slot when the health reference coverage, path integrity, and directional consistency are all satisfied.

[0067] For example, if path p has a memory value of 2 after the previous running cycle, and the current running cycle is adjacent to the previous running cycle and has the same direction, then... A value of 0.8 is acceptable. When the current coverage confidence is 0.9, the path integrity status is 1, and the direction consistency status is 1, the current path memory value is formed by the historical memory entries and the entries written in this cycle. If the current path experiences a direction reversal, then... Set to 0, and no new memory of the same direction of degradation will be added in the current cycle, while the reverse record required for subsequent closed-loop verification will be retained.

[0068] The pre-stored fault mode trajectory is established by the residual path, stage migration order and corresponding fault state in the confirmed operation record. Each fault mode trajectory consists of several allowed path nodes and necessary migration relationships. When updating the candidate fault mode, instead of using a single residual amplitude as the matching condition, the current fault evolution chain is compared with the pre-stored fault mode trajectory in terms of residual type combination, direction combination, stage start point, stage migration order and cross-cycle persistence state, and multiple candidate fault modes with similar matching degree are retained.

[0069] For candidate fault modes of increased mechanical resistance, the pre-stored trajectory may include a combination of negative residual speed following during startup, positive residual drive current, and positive residual control compensation migrating to the acceleration phase. For candidate fault modes of abnormal commutation, the pre-stored trajectory may include a combination of commutation timing residual appearing first and being accompanied by the expansion of drive current residual in adjacent operating cycles. The trajectory is only used as a data matching structure. The candidate fault modes still need to be written into the health status sequence after replay anchor point verification and cross-cycle closed-loop verification.

[0070] Furthermore, for each residual sequence written into the long-term degenerate memory state, a replay anchor point is selected based on the neighborhood relationship between the current operating condition description vector and the historical healthy operating samples; keeping the current actual response unchanged, the first residual sequence and the second residual sequence are formed by taking the counterfactual healthy response generated by the conditional baseline model and the healthy response corresponding to the replay anchor point as references; based on the consistency relationship between the first residual sequence and the second residual sequence in residual direction, generation order and cross-cycle continuity, it is determined whether the corresponding residual sequence is retained in the long-term degenerate memory state or regressed to the short-term memory state.

[0071] The replay anchor point is selected from the healthy operating samples that have passed the bidirectional consistency check. The selection criteria include having the same stage arrangement as the current operating condition, covering all stages of the current residual path, the operating condition being in the neighboring area of ​​the candidate sample set and not being the sole dominant sample of the current counterfactual health response. If there is no healthy operating sample that meets the criteria in the current candidate sample set, a healthy operating cycle with the same control intent and complete stage record is selected from the adjacent historical operating condition area and marked as the extended replay anchor point.

[0072] The first residual sequence is obtained by calculating the difference between the current actual response and the counterfactual healthy response generated by the conditional baseline model. The second residual sequence is obtained by calculating the difference between the same current actual response and the healthy response in the replay anchor point. The two residual sequences use the same stage process coordinates, the same residual type, and the same comparable interval. If a certain residual appears only in the first residual sequence but in the opposite direction in the second residual sequence, a reference sensitivity indicator is recorded. If the direction and generation order of the two residual sequences are consistent, a reference stability indicator is recorded.

[0073] The consistency state between the first residual sequence and the second residual sequence is calculated according to the following formula:

[0074] ; in, This indicates the replay consistency state in the nth running cycle. This indicates the degree of consistency in the directions of the same residual type in two residual sequences. This indicates the degree of consistency in the order in which the residual nodes are generated in two residual sequences. This indicates the degree of continuity of the corresponding residual path in adjacent running cycles. , and The coefficients represent the combination coefficients of direction, order, and continuous state in the playback verification, respectively. The plus sign indicates the weighted combination of the three types of consistent states. The coefficient constraints on the right side of the equation are used to keep the combined states within a uniform comparison range.

[0075] when Take 0.4, Take 0.35, When the value is 0.25, and the direction consistency is 1, the generation order consistency is 0.8, and the cross-cycle continuity is 0.6, Determined by three types of states, if the residual direction remains consistent but the order of generation changes, The sequence of residuals is reduced and the order difference records are retained. If the direction and order are consistent and the path appears continuously in adjacent periods, the corresponding residual sequence is retained in the long-term degenerate memory state. If the direction is reversed or the order difference persists, the corresponding residual sequence is regressed to the short-term memory state.

[0076] The residual sequence that is rolled back to the short-term memory state retains the original running cycle, counterfactual health response, replay anchor point, first residual sequence, second residual sequence, and rollback reason. After the same operating condition neighborhood is formed in subsequent running cycles, the replay verification is performed again. The rollback record does not participate in the calculation of the duration of the current fault evolution chain, but it can be used as a data basis for judging whether similar anomalies are repeatedly affected by changes in health reference, thus avoiding the direct deletion of observed operating anomalies due to a single replay result.

[0077] In a preferred embodiment, a cross-cycle closed-loop verification is performed on the residual sequence retained in the long-term degradation memory state. The first residual sequence and the second residual sequence are reverse mapped to the corresponding historical operating cycles, and the operating condition description vector, counterfactual health response and memory update record in each historical operating cycle are read. For historical operating cycles that show residual direction reversal or sequence breakage after reverse mapping, the corresponding long-term degradation memory write is cancelled. For historical operating cycles that maintain continuity, the write is retained. Based on the verified long-term degradation memory state, a motor health state sequence and a candidate fault mode sequence are generated.

[0078] The reverse mapping uses the set of occurrence cycles in the path memory slot as an index to read the stage process coordinates, candidate healthy running samples, contribution identifiers and residual association graphs saved at each path write. The currently confirmed residual node combination is projected sequentially to the same stage and the same process interval of the corresponding historical running cycle. If the direction of the corresponding residual node in the historical cycle is opposite to the direction of the current fault evolution chain, or if the node generation order in the historical cycle cannot form the current migration relationship, a revocation identifier is written into the memory update record of that historical cycle.

[0079] The undo process uses a method of recording the reverse operation without overwriting the original running data. The program reads the write amount, write cycle and memory value before update of the original path memory slot, removes the corresponding write amount from the current memory value, and keeps the residual sequence and replay verification data of the historical cycle available for query. If a migration connection in the fault evolution chain loses a continuous preceding path after undoing, the migration connection is released. After the migration connection is released, each independent path memory slot is still retained, and the path break is not equated with the non-existence of the original residual.

[0080] For continuous historical operating cycles, cross-cycle closed-loop verification retains the original long-term degradation memory and re-compares the counterfactual health response coverage status in the historical cycle with the current operating condition neighborhood. If the newly added health operation sample changes the historical counterfactual health response, the program uses the new sample to regenerate the historical counterfactual health response and recalculate the corresponding residual. When the recalculation result is consistent with the original residual in direction and order, it is retained. When the recalculation result has a reversed direction or a broken path, it is undoed.

[0081] The motor health status sequence is output according to the operating cycle. The health status record of each operating cycle includes the health status level, the path memory slot involved in the judgment, the fault evolution chain position, the coverage confidence, the playback consistency status, and the closed-loop verification status. The health status level is determined by the long-term degradation memory status retained by the closed-loop verification. The residuals in the isolated memory area or the short-term memory status do not directly improve the health status level, but the pending confirmation mark is saved in the health status record.

[0082] The candidate fault mode sequence and the motor health state sequence use the same running cycle index. Each running cycle can store multiple candidate fault modes and their corresponding fault evolution chains. The arrangement of candidate fault modes is determined based on the structural matching status of the current fault evolution chain with the pre-stored fault mode trajectory, path memory value, and closed-loop verification result. When the necessary migration relationship corresponding to a certain candidate fault mode is broken due to the revocation of historical write, the program reduces its arrangement position or removes it from the current candidate set.

[0083] In this embodiment, the training data, healthy operating sample set, boundary sample set, isolation memory area, path memory slot and pre-stored fault mode trajectory of the conditional baseline model are all managed with version identifiers. A new version record is generated each time a healthy operating sample is added, a boundary sample is transferred to the candidate sample set, an isolation memory is released or a historical memory is written and revoked. The current health status sequence is associated with the corresponding version identifier so that the counterfactual health response and residual calculation process can be restored according to the same version after the basic data changes.

[0084] Preferably, the update of healthy operation samples does not directly adopt the single operation cycle that is currently judged to be healthy. Instead, it requires that the corresponding operation cycle has not formed a residual evolution chain of the same origin in the subsequent operation cycles, and that the residual direction remains stable after performing replay anchor point verification. Operation cycles that meet the conditions are added to the set to be added. The operating condition description vector in the set to be added is checked for neighborhood duplication with the existing healthy operation samples. When the duplication degree is high, only the samples with more complete response trajectories are retained. When there are gaps in the operating condition neighborhood, the samples that can fill the gaps are retained.

[0085] In a preferred embodiment, the pre-stored fault mode trajectory can be generated from the confirmed fault operation record. The program arranges the residual association diagram in the confirmed record according to the operation cycle, extracts the repeated residual node combinations and stage migration relationships, and merges the records with the same confirmed fault type and the same migration order into a type of trajectory template. The trajectory template saves necessary nodes, optional nodes, allowed missing stages and migration direction, and does not set all residuals in a single operation as necessary nodes.

[0086] When the current running cycle has not yet ended, the running monitoring program can generate local condition description vectors and local response description vectors for the completed stages. The conditional baseline model only retrieves healthy running samples that can cover the completed stages and generates local counterfactual healthy responses. Local residuals are written into short-term memory. After the running cycle ends, the program supplements the data of subsequent stages and reconstructs the complete residual correlation graph. When the direction of local residuals is consistent with that of complete residuals, cross-cycle updates are entered. When local residuals lose continuous paths due to recovery of subsequent stages, they remain as local records within the current cycle.

[0087] In the event of a protection action during the operating cycle, the data before the protection action is processed as incomplete operation segments. The control commands and feedback data after the protection action are saved separately as protection status records and are not used together with the normal operation phase to generate the operating condition description vector. If a continuous residual path has been formed before the protection action, the path enters the isolation memory area and is accompanied by a protection truncation reason. Subsequently, the corresponding isolation record is released only when the same residual path appears in a complete operating cycle with no protection action and the operating condition neighborhood is close.

[0088] For operating cycles where the power supply voltage state changes continuously, the operating condition description vector retains the center state and fluctuation pattern of the power supply voltage. The conditional baseline model prioritizes searching for healthy operating samples with similar power supply states. If there are insufficient samples with similar power supply states, the coverage reliability of the current counterfactual health response is reduced and the writing of long-term degradation memory is restricted. This allows the determination of whether the increase in drive current and the decrease in speed originate from changes in the power supply background to be verified by the same operating condition health reference, rather than directly forming fault records based on the original amplitude.

[0089] For frequent gear shifting operation cycles, the stage connection identifier stores the stage type and speed command change direction before and after each gear shift. The response reconstruction layer generates counterfactual healthy responses for each gear interval. The residual association graph establishes migration edges at the gear shift boundary but does not directly merge the shift transient and steady-speed residuals. Only when the recovery stage after the shift continuously shows residuals in the same direction as the previous stage, are the corresponding nodes connected as cross-stage paths.

[0090] All data processing in this embodiment uses the runtime identifier, stage identifier, and stage process coordinates as common indices. The operating condition description vector, response description vector, counterfactual health response, residual description vector, residual correlation graph, and degradation memory record can be traced back to each other along the common index. Any candidate fault mode can be located to the historical runtime, residual node, and counterfactual health response source that participated in the matching, so that the data-driven prediction process maintains a definite input relationship, calculation relationship, and update relationship.

Claims

1. A data-driven method for predicting and monitoring the faults of a hair dryer motor, characterized in that, include: Acquire motor control command data, drive feedback data, and operating status data for each operating cycle of the hair dryer; A working condition description vector is constructed based on the motor control command data and the operating status data, and a response description vector is constructed based on the drive feedback data. A conditional baseline model is established between the operating condition description vector and the corresponding health response description vector using healthy operating samples, and a counterfactual health response corresponding to the current operating condition is generated. A condition-decoupled residual sequence is formed based on the actual response and the counterfactual health response; The degradation memory is updated according to two time scales: single cycle and cross cycle. The motor health status and candidate fault modes are determined based on the matching relationship between the degradation memory and the pre-stored fault mode trajectory.

2. The data-driven method for predicting and monitoring the operation of a hair dryer motor according to claim 1, characterized in that, The construction of the operating condition description vector includes: using start-up, acceleration, speed stabilization, gear shifting, deceleration and shutdown events as data segment boundaries within the operating cycle, and encoding the speed command changes, control duty cycle changes, power supply voltage fluctuations, cumulative running time and start-stop intervals in each segment into stage operating condition sub-vectors; The stage connection identifier is generated according to the sequence of each segment, and the stage condition sub-vector is combined with the stage connection identifier to form a condition description vector that maintains the order of the operation process.

3. The data-driven method for predicting and monitoring the faults of a hair dryer motor according to claim 1, characterized in that, The conditional baseline model includes a condition neighborhood retrieval layer and a response reconstruction layer. The condition neighborhood retrieval layer filters the healthy operation samples according to the consistency of control intent, operation stage, and power supply status among the condition description vectors. The response reconstruction layer reconstructs the counterfactual health response under the current condition based on the response description vectors of the filtered healthy operation samples, and retains the contribution identifier of each healthy operation sample to the counterfactual health response.

4. The data-driven method for predicting and monitoring the faults of a hair dryer motor according to claim 3, characterized in that, The decoupling residual sequence under the operating conditions includes speed following residual, drive current residual, control compensation residual, commutation timing residual and shutdown attenuation residual; A residual association graph is established according to the operating stage, generation order and contribution identifier of each residual. Residuals that appear consecutively within the same operating cycle are encoded as short-term memory states, and residuals that maintain the same direction and repeat the associated path across operating cycles are encoded as long-term degenerate memory states.

5. The data-driven method for predicting and monitoring the faults of a hair dryer motor according to claim 2, characterized in that, The generation of the stage condition sub-vector includes: mapping similar operation segments with different durations to a unified stage process coordinate system; Extract the relative time shift, local slope inflection, and compensation amount pullback position between the control command and the drive feedback on the unified stage process coordinates; An incomplete identifier is set for the operating segment that is interrupted by gear switching or protection action, and the incomplete identifier and the interruption position are retained in the operating condition description vector, without replacing the interrupted part with supplementary data.

6. The data-driven method for predicting and monitoring the faults of a hair dryer motor according to claim 3, characterized in that, The working condition neighborhood retrieval layer performs bidirectional consistency verification on the healthy operation samples. First, it retrieves candidate healthy operation samples from the current working condition description vector, and then retrieves historical working condition description vectors adjacent to the current operating cycle from each candidate healthy operation sample. Only candidate healthy running samples that satisfy the consistency of the running stage arrangement in both forward and reverse retrieval are input into the response reconstruction layer, and the remaining candidate healthy running samples are recorded as boundary samples.

7. The data-driven method for predicting and monitoring the faults of a hair dryer motor according to claim 4, characterized in that, The coverage confidence of the current counterfactual health response is calculated based on the contribution identifier, and the coverage confidence is combined with the path integrity of the residual association graph to form a memory update weight. When the healthy operation samples are concentrated in a single operating condition area or when there is a stage missing in the residual correlation path, the corresponding residuals are written into the isolated memory area. Only after the working condition neighborhood expands or the associated path is completed in a subsequent operating cycle will the residual in the isolated memory region be incorporated into the long-term degraded memory state.

8. The data-driven method for predicting and monitoring the operation of a hair dryer motor according to claim 7, characterized in that, The long-term degenerate memory state is set up with path memory slots according to the paths in the residual association graph. Each path memory slot records the occurrence period, directional consistency and stage migration order of the corresponding residual combination. When the same residual first appears in the startup phase and migrates to the acceleration or steady-state phase in subsequent operating cycles, the path memory slots before and after the migration are connected to form a fault evolution chain, and the candidate fault mode is updated according to the correspondence between different fault evolution chains and candidate fault modes.

9. The data-driven method for predicting and monitoring the faults of a hair dryer motor according to claim 8, characterized in that, For each residual sequence written into the long-term degraded memory state, a playback anchor point is selected based on the neighborhood relationship between the current operating condition description vector and the historical healthy operation sample; Keeping the current actual response unchanged, the first residual sequence and the second residual sequence are formed by taking the counterfactual health response generated by the conditional baseline model and the health response corresponding to the replay anchor point as references, respectively. Based on the consistency between the first residual sequence and the second residual sequence in terms of residual direction, generation order, and cross-period continuity, it is determined whether the corresponding residual sequence is retained in the long-term degenerate memory state or regressed to the short-term memory state.

10. The data-driven method for predicting and monitoring the operation of a hair dryer motor according to claim 9, characterized in that, Perform cross-cycle closed-loop verification on the residual sequence retained in the long-term degenerate memory state, reverse map the first residual sequence and the second residual sequence to the corresponding historical operating cycle, and read the operating condition description vector, counterfactual health response and memory update record in each historical operating cycle; For historical operating cycles that show residual direction reversal or sequence breakage after reverse mapping, the corresponding long-term degradation memory write is cancelled. For historical operating cycles that maintain continuity, the write is retained. Based on the verified long-term degradation memory state, a motor health state sequence and a candidate fault mode sequence are generated.

Images