Sensor-based permanent magnet motorized roller performance testing system

Abstract

The application discloses a sensor-based permanent magnet electric roller performance test system, and relates to the technical field of electric roller performance test, comprising a data acquisition module, a feature extraction and modeling module, a state discrimination module, a atlas matching analysis module and a health evaluation and early warning module; through multi-source real-time sensors, current, voltage, temperature, rotating speed and vibration parameters are collected to construct a roller operation data matrix; multi-dimensional feature vector groups are extracted and a nonlinear coupling state mapping model is constructed; an operation deviation index is calculated and historical working condition atlas is matched to generate a full-parameter dynamic working condition fingerprint atlas; finally, according to the evaluation results and the deviation degree, the roller health state grade and early warning information are output; the system can realize accurate identification and trend analysis of the roller operation state under multiple working conditions, and has high reliability, intelligence and application promotion value.

Description

Technical Field

[0001] This invention relates to the field of electric roller performance testing technology, and more specifically to a sensor-based permanent magnet electric roller performance testing system. Background Technology

[0002] Electric rollers are drive devices that enclose the electric motor and transmission mechanism inside the roller shell. They are widely used in equipment such as conveyors. Among them, permanent magnet electric rollers have attracted much attention due to their high efficiency, high power density, and excellent control performance. However, in actual operation, the working state of permanent magnet electric rollers is affected by many factors such as temperature, load, current, voltage, and speed. Especially under high-frequency variable loads or complex working conditions, problems such as decreased efficiency, abnormal temperature rise, and torque fluctuations may occur, affecting their service life and system stability.

[0003] Currently, the performance testing of permanent magnet electric rollers mainly relies on intermittent manual testing or the collection of only a single parameter. This results in poor testing accuracy and real-time performance, making it difficult to achieve multi-dimensional and multi-state dynamic monitoring and analysis, and thus hindering a comprehensive evaluation of their operational performance. Therefore, there is an urgent need for a performance testing system that can integrate signals from multiple sensors, acquire key operating parameters of permanent magnet electric rollers in real time, and achieve intelligent and automated evaluation and data visualization processing. This is a crucial technical problem that needs to be solved to improve roller operation safety and maintenance efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a sensor-based permanent magnet electric roller performance testing system to address the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a sensor-based permanent magnet electric roller performance testing system, comprising:

[0006] The data acquisition module collects multi-source real-time operating parameters of the permanent magnet electric drum under set operating conditions, including current, voltage, temperature, speed and vibration signals, and constructs a multi-dimensional operating data matrix of the drum.

[0007] The modeling module extracts a multidimensional feature vector set of the drum operation based on the time-series change trend of each parameter in the data matrix and the nonlinear coupling relationship between the parameters, and constructs a drum performance evaluation model. The model uses a nonlinear coupled state mapping function to establish the mapping relationship between the parameter space and the state space.

[0008] The state discrimination module compares and analyzes the roller performance evaluation model with the preset performance index boundary range of the roller to generate a roller operation deviation index, which is used to determine whether there is an abnormal deviation in the current operating state of the roller.

[0009] The graph matching and analysis module performs pattern matching between the feature vector group of the current operating status and the working condition feature graph of the corresponding roller model in the historical roller feature database to form a full-parameter dynamic working condition fingerprint graph and output a performance evaluation result set.

[0010] The health assessment and early warning module calculates the health status level of the rollers based on the assessment result set and the roller operation deviation index, and outputs status early warning information.

[0011] Preferably, in the modeling module, the extraction of the multidimensional feature vector set of the roller operation includes:

[0012] A parameter response surface plot is constructed based on a multidimensional operational data matrix to identify the nonlinear evolution trajectory of each parameter over time.

[0013] By using the roller condition transformation tensor, data slices from different time periods are constructed into a third-order tensor array;

[0014] The local weighted reconstruction method based on the parameter coupling tension coefficient performs dimensionality recompression on the highly correlated parameter group to highlight the feature contribution value;

[0015] The compressed feature subspace is normalized and phase aligned to generate a multidimensional feature vector set for roller performance evaluation modeling.

[0016] Preferably, the construction of the drum performance evaluation model includes:

[0017] Based on the feature vector set of drum operation, a multivariable state evolution trajectory in the parameter space is constructed.

[0018] Define a typical roller working condition boundary cluster and map the physical parameter range corresponding to each feature dimension to a continuous interval in the state space;

[0019] A nonlinear coupled state mapping function is established through a differentiable state transformation network to describe the joint driving effect of multiple parameter inputs on the drum state.

[0020] Preferably, the method for generating the drum running deviation index is as follows: extracting a normalized feature vector from the running data: Where d is the feature dimension and T is the vector transpose;

[0021] Using historical normal operation data of the drum Fit a Gaussian mixture model, where N is the total number of data points. Assume the model contains K mixture components. Then the joint probability density function is... for: In the formula, This represents the weight of the k-th Gaussian component. This represents the mean vector of the k-th component. Let the covariance matrix of the k-th component be denoted as . Given a multivariate Gaussian density function; calculate the probability density of the current eigenvector. The expression is: ; This represents the currently collected feature vector of the drum's movement; the drum's movement deviation index is calculated, expressed as: In the formula, The index represents the deviation of the drum's movement.

[0022] Preferably, the spectral matching analysis module includes:

[0023] A hierarchical indexing system for roller models was constructed, and the historical feature database was coded with multidimensional tags according to roller structural parameters, power level and typical working condition type.

[0024] A phase correlation comparison method is used to perform time alignment processing on the current feature vector group and the candidate working condition map.

[0025] By using a dynamic cosine similarity matrix, the local similarity curves between the current working condition and the historical working condition under multiple feature dimensions are calculated to form a similarity heatmap.

[0026] A dynamic working condition fingerprint map with full parameters is generated based on the similarity distribution pattern, and a performance evaluation result set is output in combination with a set threshold.

[0027] Preferably, the generation of a full-parameter dynamic operating condition fingerprint map and the output of a performance evaluation result set include:

[0028] Based on the similarity heatmap between the current feature vector group and the historical operating condition map, a multi-dimensional similarity response surface is constructed, and the principal axis curve of the response shape is extracted as the fingerprint structure baseline.

[0029] The concave regions of the similarity response surface are locally amplified to highlight potential atypical offset behaviors.

[0030] The extracted response patterns are mapped to the standard working condition pattern template corresponding to the roller model, and the final full-parameter dynamic working condition fingerprint pattern is formed through spatial distortion matching.

[0031] Based on the established hierarchical discrimination threshold system, the overall deviation of the fingerprint spectrum is scored and a performance evaluation result set is output, including the current operating condition level, evolution trend and prediction suggestions.

[0032] Preferably, the calculation of the drum health status level includes:

[0033] The performance score of the drum under current operating conditions ∈[0,1] is combined with the running deviation index DRDI to construct a two-dimensional evaluation point, and then projected onto the set health level distribution interval, with the health level set to level five;

[0034] The performance scores of the previous three time windows in the current period are collected, and the smoothed average score is calculated using the first-order exponential smoothing method. The expression is: ;in These are the scores for the first three periods, respectively;

[0035] Construct a residual function between the DRDI and the scoring curve, and calculate the residual value between the current state and the curve. : Where f(DRDI) is the regression curve of the score and deviation index in the historical normal sample;

[0036] according to The three parameters consisting of DRDI and residual value R are scored using the set health level decision rules, outputting the health level and corresponding color code, and generating roller maintenance suggestions.

[0037] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0038] 1. This invention, by constructing a multi-dimensional health status discrimination mechanism that combines a drum performance evaluation model with a deviation index, enables real-time quantitative analysis of the operating status of permanent magnet electric drums under complex working conditions. Compared with traditional methods that rely on a single threshold or static rules, this invention can dynamically identify potential weak anomalies, operating condition drift, and latent degradation behaviors, improving diagnostic sensitivity and early warning accuracy, and effectively reducing the risk of sudden drum failures.

[0039] 2. This invention introduces a similarity heatmap matching algorithm between the dynamic operating condition fingerprint map and the historical operating condition map, enabling visualized identification and trend tracking of the drum's operating status across multiple operating conditions, dimensions, and cycles. By combining scoring smoothing, deviation residual analysis, and a multi-level health level output mechanism, this system can support intelligent maintenance scheduling and lifecycle management of the drum. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0041] Figure 1 This is a flowchart of the system modules of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] For examples, please refer to Figure 1 As shown, the sensor-based permanent magnet electric roller performance testing system described in this embodiment includes:

[0044] The data acquisition module collects multi-source real-time operating parameters of the permanent magnet electric drum under set operating conditions, including current, voltage, temperature, speed and vibration signals, and constructs a multi-dimensional operating data matrix of the drum.

[0045] The modeling module extracts a multidimensional feature vector set of the drum operation based on the time-series change trend of each parameter in the data matrix and the nonlinear coupling relationship between the parameters, and constructs a drum performance evaluation model. The model uses a nonlinear coupled state mapping function to establish the mapping relationship between the parameter space and the state space.

[0046] The state discrimination module compares and analyzes the roller performance evaluation model with the preset performance index boundary range of the roller to generate a roller operation deviation index, which is used to determine whether there is an abnormal deviation in the current operating state of the roller.

[0047] The graph matching and analysis module performs pattern matching between the feature vector group of the current operating status and the working condition feature graph of the corresponding roller model in the historical roller feature database to form a full-parameter dynamic working condition fingerprint graph and output a performance evaluation result set.

[0048] The health assessment and early warning module calculates the health status level of the rollers based on the assessment result set and the roller operation deviation index, and outputs status early warning information.

[0049] In this embodiment, a multi-source data acquisition scheme for performance testing of permanent magnet electric rollers is provided, and the specific process is as follows:

[0050] First, before the permanent magnet electric drum is put into operation, basic sampling parameters are set according to the drum's designed speed range and load level. The system introduces a dynamic sampling frequency adjustment mechanism, that is, a microcontroller unit (MCU) is configured at the sensor input front end, which can adjust the sampling frequency in real time according to the speed acceleration and load fluctuations. For example, when the drum enters the heavy load start-up state, the system increases the sampling frequency of current, voltage and vibration signals to 5kHz to capture transient fluctuation characteristics, while automatically reducing the frequency to 1kHz during steady-state operation to reduce redundant data.

[0051] Secondly, current and voltage parameters are acquired by tension-isolated sensors configured on the power supply line. These sensors use magnetic coupling to sense electromagnetic changes in the main power supply line, thereby avoiding electrical interference with the original control system and ensuring that the test system can operate independently without changing the drum control circuit.

[0052] Temperature data acquisition employs a distributed thermocouple network structure, where several thermocouples (such as type K) are attached or embedded on the surface of the drum shell and the bearing area to form an axial and radial multi-point temperature monitoring array. The drum thermal field distribution is reconstructed using the differential interpolation method.

[0053] Furthermore, since the acquired signals come from asynchronous sampling channels (with different sampling rates for temperature, vibration, current, etc.), they need to undergo time-domain alignment processing via the data acquisition module. This module employs a unified timestamp management and interpolation mechanism to ensure that different signals have corresponding time labels within the same sampling period. Ultimately, it constructs a multi-dimensional operational data matrix with high time consistency that can be directly input into the state assessment model, laying the foundation for subsequent performance analysis.

[0054] The state assessment model of this invention is based on the construction of a multi-dimensional feature vector group of the drum, and establishes a nonlinear coupling mapping between the "parameter space" and the "state space" through a "differentiable state transformation network". The modeling process includes: extracting multi-dimensional parameter responses from time-series data, constructing a third-order tensor array of "time × parameter dimension × working condition number"; using a deep neural network with continuous differentiability (such as a feedforward network with ReLU activation function) to learn the state mapping; using historical normal working condition data as training samples, fitting the nonlinear relationship between the multi-parameter joint input and the state response, and outputting the drum operation deviation index.

[0055] In this embodiment, based on the multidimensional operation data matrix of the drum, the operation feature vector extraction and performance evaluation model construction are carried out, specifically including the following steps:

[0056] First, based on the operating data collected from the drum under different loads and startup phases, a parametric response surface plot was constructed. Key parameters such as time, current, voltage, temperature, and vibration were transformed into a three-dimensional response surface to analyze the nonlinear characteristics of each parameter changing over time. During the observation of the surface changes, a significant coupling delay effect was found between the vibration parameters and the temperature rise rate under different operating conditions.

[0057] Subsequently, to capture the coupled temporal behavior between parameters, a third-order tensor representation method was used to model the runtime data. The specific method is as follows:

[0058] The runtime data for different time periods is sliced ​​into a "parameter × time" matrix;

[0059] Stack them in chronological order to form a third-order tensor;

[0060] A "coupling tension coefficient matrix" is constructed by analyzing the rate of change of local covariance among the various parameter dimensions to reflect the time delay coupling relationship between the parameters.

[0061] Based on this tensor structure, coupling delays can be identified, such as a covariance jump 0.5s after the temperature rise due to changes in vibration intensity.

[0062] Next, we move on to the model building phase. First, based on the multi-dimensional feature vectors extracted above, we use historical working condition samples of the drum to set a set of typical working condition boundary clusters, dividing each feature dimension into "normal," "critical," and "abnormal" state intervals. Then, we construct a class of differentiable state transformation function networks to simulate the nonlinear joint effects of multiple physical parameters under multiple working conditions, thus establishing a mapping relationship from the "parameter space" to the "state space."

[0063] Finally, to ensure model accuracy, residual verification of the model output is performed using experimental calibration data. This involves analyzing the differences between the model's predicted state and the measured state point by point, and adjusting the model's boundary response using a deviation weighting mechanism, thereby forming a complete performance evaluation model applicable to multiple working conditions of the drum.

[0064] The technical means shown in this embodiment ensure that the evaluation model has stability, interpretability and generalization ability under multi-parameter input conditions, providing an accurate basis for subsequent state judgment and fault prediction.

[0065] In the state discrimination module, normalized feature vectors are extracted from the running data: Where d is the feature dimension, such as temperature gradient, root mean square value of vibration, current fluctuation rate, etc., and T is the vector transpose.

[0066] Using historical normal operation data of the drum To fit a GMM model, where N is the total number of data points, and the model contains K mixture components, the joint probability density function is: In the formula, This represents the weight of the k-th Gaussian component. This represents the mean vector of the k-th component. Let the covariance matrix of the k-th component be denoted as . Given a multivariate Gaussian density function; calculate the probability density of the current eigenvector. The expression is: ;

[0067] in This represents the currently collected feature vector of the drum's operation. Its dimension matches the input dimension in the training model, and it includes various normalized parameters such as temperature, current, voltage, and vibration. It is used to evaluate the probability density of the current state occurring in the training model to determine whether it is a normal operating state. A higher value indicates that the current state is close to the normal region in the trained model; if A lower value indicates that the current state may be abnormal. The roller running deviation index is calculated using the following expression: In the formula, The index represents the deviation of the drum's movement.

[0068] If DRDI≈0, it indicates that the state is close to normal; if This indicates that the state deviates significantly from the model density center, i.e., a potential anomaly.

[0069] In this embodiment, for a certain model (rated power 11kW, belt diameter 315mm) of permanent magnet electric drum, its historical operating data under multiple typical working conditions (no-load start, full-load operation, high temperature environment) were collected, including five dimensions such as current RMS value, voltage fluctuation amplitude, housing temperature rise rate, spindle speed stability and axial vibration RMS value.

[0070] When constructing the historical operating condition database, this invention adopts a multi-level classification label of "model + power level + typical operating condition"; the encoding structure is: [model_ID]_[power level]_[operating condition type], such as MGD315_11kW_FL representing full-load operating condition; the matching algorithm uses phase correlation comparison, that is, it finds the maximum phase consistency point in the frequency domain through Fourier transform, which corresponds to the optimal alignment position of the time series data on the time axis.

[0071] When the drum is running in real time, the system acquires five-dimensional feature values ​​within the current time window, forming a feature vector group. For example, the temperature rise rate in the current window is 1.6°C / min, and the vibration RMS is 2.1 mm / s. The system calls the corresponding historical operating condition map and uses a phase correlation matching algorithm to compare the matching degree of the feature sequences on the time axis. Specifically, this includes: after comparing the current operating condition with the historical map, calculating a multi-dimensional cosine similarity heatmap; projecting the heatmap to form a response principal axis curve and dividing it into 67 equal-width segments; calculating the similarity score for each segment, setting the weight of each segment, calculating the weighted average, and outputting the final performance score.

[0072] Next, the system constructs a similarity matrix and calculates the cosine similarity between the real-time and standard spectra for each feature dimension, generating a two-dimensional heatmap. For example, the average similarity between the current operating condition and the "full load operation" spectra is 0.72, but the vibration feature term is only 0.41, indicating a possible structural anomaly.

[0073] The system performs weighted projection of the similarity heatmap along the feature dimension, extracts the principal axis response curve, and performs map registration with the standard working condition map. Through the spatial distortion matching method, a five-dimensional dynamic working condition fingerprint map is formed, which records the "working condition signature" of the roller in image form.

[0074] In constructing a full-parameter dynamic operating condition fingerprint map, this invention employs the following map registration and scoring process:

[0075] Calculate the five-dimensional feature vector set of the current operating condition;

[0076] By comparing historical standard maps, a cosine similarity heatmap of "current operating conditions vs. standard maps" is constructed.

[0077] Phase correlation matching is performed on the time axis and feature axis, and the current feature curve is found to have the maximum similarity position in the standard map by interpolation and sliding window.

[0078] The response surface structures of the two are aligned using spatial warp matching methods (such as dynamic time warping (DTW) or weighted deformation functions) to complete the spectral registration.

[0079] The registered heatmap is divided into 67 equal-width segments on the time axis;

[0080] The average similarity score for each segment is calculated and denoted as... (i=1..67);

[0081] Set segment weights (By default, scores are equal or weighted according to the importance of the work conditions), the final total score is: This score is the current working condition fingerprint score (range 0~1), used for health level judgment and trend prediction.

[0082] During this process, if there are consecutive low similarity bands in the similarity matrix (such as continuously low current + temperature rise), the system automatically triggers local anomaly enhancement processing to improve the map resolution and mark suspicious areas.

[0083] The system performs an overall score on the fingerprint spectrum, and sets the status classification thresholds as follows:

[0084] Normal: Score ≥ 0.85;

[0085] Metastability: 0.70 ≤ score < 0.85;

[0086] Critical threshold: 0.55 ≤ score < 0.70;

[0087] Abnormal: Score < 0.55;

[0088] The current operating condition score is 0.61, falling into the critical zone. Due to its proximity to the score boundary, the system further introduces a fuzzy state factor (to calculate the overlap of fluctuations between features) for interval correction, and finally outputs: State level: Critical; Main deviation parameters: Vibration, temperature rise; Trend suggestion: It is recommended to clean the drum heat pipe and check the bearing; Graph record: Save the current operating condition graph into the database for subsequent trend analysis.

[0089] This invention tracks the changing trend of the score value sequence through a three-period sliding window, specifically as follows:

[0090] Record the score at the current moment And the scores from the first two periods and ;

[0091] The trend mean is generated using the first-order exponential smoothing method (EWMA): Where α∈(0,1) is the smoothing factor, and a suggested value is 0.3~0.5;

[0092] if This indicates a deteriorating trend;

[0093] If the scores fluctuate greatly but the mean tends to stabilize, the trend is considered "metastable".

[0094] This trend can be used to predict the direction of health level changes, serving as input for roller maintenance strategies.

[0095] In this embodiment, a technical solution for classifying the operational health status of permanent magnet electric drums is provided. A multi-factor health determination system is constructed by combining performance evaluation scores, deviation index (DRDI), and score residuals, which is suitable for outputting drum status levels and supporting maintenance decisions under complex working conditions.

[0096] During the drum's operation, the system calculates its overall performance score under its current operating conditions in real time. The range is 0 to 1. The scoring indicators include parameters such as temperature rise rate, current fluctuation, voltage imbalance, and vibration intensity.

[0097] Meanwhile, the system uses a trained Gaussian mixture model to calculate the operating deviation index DRDI of the current working condition, reflecting the degree of deviation of the drum condition from the historical normal state distribution.

[0098] The system is based on ( The two-dimensional input is used to construct runtime status evaluation points for subsequent level mapping.

[0099] To suppress the impact of occasional disturbances on state judgment, a first-order exponentially weighted average (EWMA) algorithm is introduced to weight the scores from the most recent three scoring periods, and the smoothed average score is calculated. The expression is: ;in These are the scores for the first three periods. This operation can effectively reduce the interference of short-term fluctuations on the assessment level.

[0100] The system uses historical health samples to establish an empirical regression curve f(DRDI) between the health score and the DRDI, and calculates the residual value between the current state and the curve. The expression is: If the residual is too large, it indicates a structural inconsistency between the score and the deviation index, which may indicate a skewed state that is "seemingly healthy but actually harbors hidden dangers".

[0101] Construction of the f(DRDI) regression function: During the system modeling phase, using a historical health condition sample set, extract the corresponding DRDI and score pairs; use the least squares method to fit a regression curve, such as a polynomial fit or a cubic spline function. Where s, b, and c are empirical regression coefficients, representing the nonlinear adjustment factor (s) for the deviation index, the linear rate of decline coefficient (b), and the baseline score under normal operating conditions (c), respectively. This empirical regression function is obtained by least-squares fitting of the DRDI and score pairs in historical samples. The function f is used to express "the expected change of the score with the deviation index under normal distribution". If the residual R > the threshold (e.g., 0.1), it indicates that the score is inconsistent with the degree of deviation, and there may be a "seemingly normal but structurally problematic" operating condition.

[0102] In this embodiment, the health status of the drum is divided into five levels, from healthy to abnormal: H0, H1, H2, H3, and H4. Each level is determined by three core indicators: the smoothed performance score. The running deviation index DRDI and the score residual R (i.e., the difference between the current score and the expected deviation value).

[0103] H0 level (completely healthy): When the average performance score of the roller is higher than 0.85, the deviation index DRDI is less than 1.2, and the score residual is less than 0.05, the system determines that the roller is in a completely healthy state with extremely high operational stability and no intervention is required.

[0104] H1 level (stable): When the score is between 0.75 and 0.85, the deviation index does not exceed 2.0, and the residual is less than 0.1, the system considers the drum condition to be basically stable and can continue to operate, but routine inspections are recommended.

[0105] H2 level (sub-health): When the score drops to the range of 0.65 to 0.75, the deviation index is less than 3.0, and the residual value is below 0.15, the system determines that the roller may have a slight abnormality, such as the early stage of bearing wear, rapid temperature rise, etc. It is recommended to arrange an inspection in the short term.

[0106] H3 level (critical anomaly): If the score is between 0.50 and 0.65, the deviation index is below 4.0, and the residual value is less than 0.2, it indicates that the drum is in a critical operating state with obvious abnormal characteristics, and pre-maintenance should be arranged as soon as possible.

[0107] H4 level (serious anomaly): When the score is below 0.50, the deviation index is greater than or equal to 4.0, and the residual between the score and the model expectation exceeds 0.2, the system determines that the drum has a serious anomaly risk, such as continuous vibration exceeding the limit, rapid temperature rise, or uncontrolled deviation of the operating conditions. It is recommended to stop the machine immediately and carry out maintenance.

[0108] The system automatically outputs the corresponding health level code (such as "H2") and color code (such as yellow) according to the above rules, and sends diagnostic suggestions and risk levels to the maintenance system through the early warning interface.

[0109] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A sensor-based permanent magnet electric roller performance testing system, characterized in that: include: The data acquisition module collects multi-source real-time operating parameters of the permanent magnet electric drum under set operating conditions, including current, voltage, temperature, speed and vibration signals, and constructs a multi-dimensional operating data matrix of the drum. The modeling module extracts a multidimensional feature vector set of the drum operation based on the time-series change trend of each parameter in the data matrix and the nonlinear coupling relationship between the parameters, and constructs a drum performance evaluation model. The model uses a nonlinear coupled state mapping function to establish the mapping relationship between the parameter space and the state space. The state discrimination module compares and analyzes the roller performance evaluation model with the preset performance index boundary range of the roller to generate a roller operation deviation index, which is used to determine whether there is an abnormal deviation in the current operating state of the roller. The graph matching and analysis module performs pattern matching between the feature vector group of the current operating status and the working condition feature graph of the corresponding roller model in the historical roller feature database to form a full-parameter dynamic working condition fingerprint graph and output a performance evaluation result set. The health assessment and early warning module calculates the health status level of the roller based on the assessment result set and the roller operation deviation index, and outputs status early warning information. The calculation of the health status level of the drum includes: The performance score of the drum under current operating conditions ∈[0,1] is combined with the running deviation index DRDI to construct a two-dimensional evaluation point, and then projected onto the set health level distribution interval, with the health level set to level five; The performance scores of the previous three time windows in the current period are collected, and the smoothed average score is calculated using the first-order exponential smoothing method. The expression is: ;in These are the scores for the first three periods, respectively; Construct a residual function between the DRDI and the scoring curve, and calculate the residual value between the current state and the curve. : Where f(DRDI) is the regression curve of the score and deviation index in the historical normal sample; according to The three parameters consisting of DRDI and residual value R are scored using the set health level decision rules, outputting the health level and corresponding color code, and generating roller maintenance suggestions.

2. The sensor-based permanent magnet electric roller performance testing system according to claim 1, characterized in that: In the modeling module, the extracted multidimensional feature vector set of the roller operation includes: A parameter response surface plot is constructed based on a multidimensional operational data matrix to identify the nonlinear evolution trajectory of each parameter over time. By using the roller condition transformation tensor, data slices from different time periods are constructed into a third-order tensor array; The local weighted reconstruction method based on the parameter coupling tension coefficient performs dimensionality recompression on the highly correlated parameter group to highlight the feature contribution value; The compressed feature subspace is normalized and phase aligned to generate a multidimensional feature vector set for roller performance evaluation modeling.

3. The sensor-based permanent magnet electric roller performance testing system according to claim 2, characterized in that: The construction of the drum performance evaluation model includes: Based on the feature vector set of drum operation, a multivariable state evolution trajectory in the parameter space is constructed. Define a typical roller working condition boundary cluster and map the physical parameter range corresponding to each feature dimension to a continuous interval in the state space; A nonlinear coupled state mapping function is established through a differentiable state transformation network to describe the joint driving effect of multiple parameter inputs on the drum state.

4. The sensor-based permanent magnet electric roller performance testing system according to claim 1, characterized in that: The method for generating the drum running deviation index is as follows: extract a normalized feature vector from the running data: Where d is the feature dimension and T is the vector transpose; Using historical normal operation data of the drum Fit a Gaussian mixture model, where N is the total number of data points. Assume the model contains K mixture components. Then the joint probability density function is... for: In the formula, This represents the weight of the k-th Gaussian component. This represents the mean vector of the k-th component. Let the covariance matrix of the k-th component be denoted as . Given a multivariate Gaussian density function; calculate the probability density of the current eigenvector. The expression is: ; This represents the currently collected feature vector of the drum's movement; the drum's movement deviation index is calculated, expressed as: In the formula, The index represents the deviation of the drum's movement.

5. The sensor-based permanent magnet electric roller performance testing system according to claim 1, characterized in that: The map matching analysis module includes: A hierarchical indexing system for roller models was constructed, and the historical feature database was coded with multidimensional tags according to roller structural parameters, power level and typical working condition type. A phase correlation comparison method is used to perform time alignment processing on the current feature vector group and the candidate working condition map. By using a dynamic cosine similarity matrix, the local similarity curves between the current working condition and the historical working condition under multiple feature dimensions are calculated to form a similarity heatmap. A dynamic working condition fingerprint map with full parameters is generated based on the similarity distribution pattern, and a performance evaluation result set is output in combination with a set threshold.

6. The sensor-based permanent magnet electric roller performance testing system according to claim 5, characterized in that: Among them, the generation The full-parameter dynamic operating condition fingerprint spectrum and output performance evaluation result set include: Based on the similarity heatmap between the current feature vector group and the historical operating condition map, a multi-dimensional similarity response surface is constructed, and the principal axis curve of the response shape is extracted as the fingerprint structure baseline. The concave regions of the similarity response surface are locally amplified to highlight potential atypical offset behaviors. The extracted response patterns are mapped to the standard working condition pattern template corresponding to the roller model, and the final full-parameter dynamic working condition fingerprint pattern is formed through spatial distortion matching. Based on the established hierarchical discrimination threshold system, the overall deviation of the fingerprint spectrum is scored and a performance evaluation result set is output, including the current operating condition level, evolution trend and prediction suggestions.

Images