Motor data anomaly detection method and system based on internet of things

Abstract

The application belongs to the technical field of data processing, and particularly relates to a motor data anomaly detection method and system based on the Internet of Things, which comprises the following steps: obtaining multi-dimensional time sequence data of a motor multi-position sensor, calculating Mahalanobis distance to divide out a normal motor group based on a median data vector formed by a median value in the data and a target data vector formed by real-time data; for the normal motor, a health response characteristic reflecting dynamic change is calculated, and a normal expected vector model is established; the consistency between actual change of a to-be-tested target vector of a to-be-tested motor and theoretical change of the to-be-tested target vector corrected to an expected vector is compared, and a to-be-tested deviation degree is calculated; finally, a to-be-tested normal index is calculated in combination with the to-be-tested deviation degree and the difference of the median vector, and the index is used for continuous monitoring and abnormal early warning. The application solves the technical problems that the prior art is susceptible to individual differences and abnormal data interference and cannot adapt to dynamic working conditions.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a method and system for detecting abnormal motor data based on the Internet of Things (IoT). Background Technology

[0002] Remote, unmanned pumping stations are widely used in municipal water supply. These stations are often located in remote areas far from towns and cities, presenting challenges such as long inspection cycles, high transportation costs, and delayed response to sudden faults. If early faults in the motor, such as bearing wear or winding short circuits, are not detected in time, it can not only cause the individual motor to stop operating but also trigger a chain reaction, causing the entire pumping station's water supply or drainage function to be interrupted. This can affect residential water use, agricultural irrigation, or sewage treatment progress. Furthermore, the escalation of the fault can worsen the motor damage, increasing the time and economic cost of subsequent repairs. In addition, during operation, the motor's condition is affected by various factors such as fluctuations in water supply load, changes in ambient temperature, and the natural aging of mechanical components. Three key parameters—temperature, current, and vibration—will change accordingly, and subtle anomalies in these parameters often serve as early warning signals of faults.

[0003] In order to overcome the time and space limitations of operation and maintenance of motor groups in remote unattended pumping stations and avoid functional interruption and economic losses due to failures, the current technology mainly collects three types of data of motor vibration, temperature and current through sensors, uses the hierarchical analysis method to determine the weight of each parameter, and then obtains the fault characteristic value by weighted summation. When the characteristic value exceeds the preset range, it is determined that the motor has a fault.

[0004] However, relying solely on comparing the parameters of a single motor with preset fault characteristic values ​​cannot distinguish between individual motor differences and actual faults, leading to healthy motors being misjudged due to fluctuations in individual parameters. Furthermore, since motor operation is affected by factors such as load fluctuations and ambient temperature, the parameters change dynamically. Existing technologies have not built dynamic models that adapt to changes in operating conditions, and relying solely on static thresholds cannot meet the requirements for fully automated detection in unattended scenarios. Summary of the Invention

[0005] To address the technical problems of existing technologies being susceptible to individual differences and abnormal data interference, and being unable to adapt to dynamic working conditions, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for detecting abnormal motor data based on the Internet of Things, comprising:

[0007] Sensor data from different positions of the motor are synchronously acquired at a fixed frequency within a preset time period and arranged chronologically to obtain multidimensional time-series data. The median value of each dimension in the multidimensional time-series data, and the data of each dimension at the same time, are denoted as the median data vector and the target data vector. Based on the relationship and difference between the median data vector and the target data vector, an initial covariance matrix is ​​obtained, and a Mahalanobis distance sequence is calculated. Based on the distance interval, anomaly demarcation points are calculated to obtain the normal motor group. The normal median vector and normal target vector of an individual motor in the normal motor group are obtained. Based on the normal target vector of an individual motor, the normal median vector and normal target vector of the individual motor are obtained. The health response characteristics are calculated based on the difference between the change range within a given time period and the deviation of the normal target vector from the previous moment. The normal expected vector is then calculated based on the influence of the health response characteristics on the correction process of the normal target vector. The target vector and expected vector for any motor under test are obtained. The deviation is calculated based on the consistency between the actual change of the target vector and the theoretical change that needs to be corrected to the expected vector. The normal index is calculated based on the influence of the deviation on the target vector and the difference between the median vector and the target vector. Continuous monitoring and abnormal warnings are then provided.

[0008] This invention addresses the shortcomings of existing technologies, such as the lack of a normal benchmark based on a group of pump station motors, adaptability to changing operating conditions, and the elimination of abnormal interference, as well as the need for manual intervention and incomplete judgment. This invention establishes comprehensive and time-aligned multidimensional time-series data by synchronously collecting data at fixed frequencies from key parts of the motors, providing an accurate data foundation for subsequent analysis. Then, by generating a median data vector as the normal benchmark for the group and a target data vector as a real-time reference for individual motors, and combining the covariance matrix and Mahalanobis distance method, the influence of data correlation and magnitude differences are eliminated, making deviation calculation more accurate. Subsequently, the anomaly boundary point is automatically identified to filter normal motor groups, establishing a pure and reliable normal benchmark and avoiding interference from abnormal data. Next, based on the operating patterns of healthy motors in various dimensions, a normal expectation vector that can meet the changing operating conditions is generated. Finally, by combining the measured deviation and the normal index, anomalies are comprehensively judged from both the aspects of consistency of change and inherent stability. The entire process requires no manual intervention.

[0009] Preferably, obtaining multidimensional time-series data includes:

[0010] A triaxial vibration sensor is installed on the bearing end cover of the motor, a current sensor is connected in series in the power supply line of the motor, and a temperature sensor is attached to the surface of the motor housing near the stator. Temperature data, current data, and vibration data from the sensors at different positions of the motor are acquired synchronously at akHz within a preset time period, and time-aligned. The data are then arranged in order according to the timestamps to form multidimensional time-series data, which includes temperature data sequences, current data sequences, and vibration data sequences.

[0011] Preferably, obtaining multidimensional time-series data includes:

[0012] A triaxial vibration sensor is installed on the bearing end cover of the motor, a current sensor is connected in series in the power supply line of the motor, and a temperature sensor is attached to the surface of the motor housing near the stator. Temperature data, current data, and vibration data from the sensors at different positions of the motor are acquired synchronously at akHz within a preset time period, and time-aligned. The data are then arranged in order according to the timestamps to form multidimensional time-series data, which includes temperature data sequences, current data sequences, and vibration data sequences.

[0013] Preferably, calculating and obtaining the Mahalanobis distance sequence includes:

[0014] Using the standard sample covariance matrix formula, the initial covariance matrix is ​​calculated for all data in the target time series data set composed of all target data vectors. Based on the initial covariance matrix, the Mahalanobis distance method is used to calculate the distance between all target data vectors and the median data vector in the multidimensional time series data, which is denoted as Mahalanobis distance. The Mahalanobis distance of all motors is calculated, and all the obtained Mahalanobis distances are arranged from smallest to largest to obtain the Mahalanobis distance sequence.

[0015] This invention, when analyzing motor data, uses a standard method to calculate the initial covariance matrix. This fully considers the inherent relationship between temperature, current, and vibration, avoiding isolated analysis of any single dimension and preventing judgment biases caused by neglecting the correlation between data. Based on this matrix, Mahalanobis distance is then used to calculate distances, eliminating the influence of different data magnitudes and ensuring that the calculated distances more accurately reflect the degree to which the motor deviates from its normal state. Arranging the distances of all motors sequentially provides a clear visual representation of the different deviations from normal states, offering a clear basis for subsequently identifying the boundary between normal and abnormal motors and further reducing the possibility of misjudgment.

[0016] Preferably, calculating the abnormal boundary point to obtain the normal motor group includes:

[0017] The difference between all adjacent Mahalanobis distance values ​​in the Mahalanobis distance sequence is calculated to obtain multiple difference values. Among the multiple difference values, the largest difference value is determined, and the position point in the Mahalanobis distance sequence corresponding to the largest difference value is determined as the abnormal boundary point. The order i of the abnormal boundary point in the Mahalanobis distance sequence is obtained, and the points after the i-th gradient point in the Mahalanobis distance sequence are removed. The remaining gradient points are recorded as the standard point set, and the motors corresponding to all points in the standard point set are recorded as the normal motor group.

[0018] This invention, when screening normal motor groups, calculates the difference between adjacent values ​​in the Mahalanobis distance sequence and then locates the position corresponding to the maximum difference as the anomaly boundary point. The calculation process is entirely automated, eliminating the need for manually setting judgment criteria, thus perfectly meeting the requirement of remote, unattended pumping stations that do not require human intervention. Motors following the boundary point are removed, resulting in a normal motor group. This ensures that the data in this group conforms to normal operating characteristics, preventing the inclusion of data from abnormal motors and avoiding interference from anomalous data in subsequent analysis and judgment. This provides a clean data foundation for establishing a reliable benchmark for normal operation.

[0019] Preferably, the health response characteristics satisfy the following expression:

[0020] ;

[0021] In the formula, This represents the health response characteristics of a single motor at time t in the multidimensional time-series data of a normal multidimensional dataset. , This represents the normal target vector at time t and time t-1 in the multidimensional time series data of a single motor in a normal multidimensional dataset; The normal median vector represents the multidimensional time-series data of a single motor in a normal multidimensional dataset.

[0022] This invention calculates health response characteristics that clearly reveal the adjustment habits of a healthy motor in different aspects. For example, when faced with deviations in temperature, current, or vibration from normal conditions, it shows the proportion of the actual change in the healthy motor per unit time to the theoretically required adjustment. This intuitively reflects the response speed and self-correction phenomena of the healthy motor in different dimensions. Unlike traditional methods that judge all dimensions based on a single indicator, this invention can accurately uncover the unique operating patterns of a healthy motor in each dimension, providing crucial evidence for subsequent predictions of the motor's expected state and making subsequent assessments of whether the motor under test is abnormal more targeted.

[0023] Preferably, calculating the normal expectation vector includes:

[0024] Based on the normal target vector and health response characteristics at time t in the multidimensional time series data of a single motor in the normal multidimensional dataset, the difference between the normal target vector at time t and the normal median vector is calculated and denoted as the instantaneous regression driving vector. The health response characteristics of the multidimensional time series data of a single motor at time t are multiplied with the instantaneous regression driving vector to obtain the expected change. The normal target vector of a single motor at time t is added with the expected change to obtain the normal expected vector of the single motor at time t+1.

[0025] Preferably, the deviation to be measured satisfies the following expression:

[0026] ;

[0027] In the formula, This represents the measured deviation of the motor under test at time t within a preset time period; , Let represent the target vector to be measured at time t and time t+1; This represents the expected vector to be measured at time t+1. This is the modulus symbol.

[0028] This invention, when determining whether the changes in a motor under test are normal, calculates the deviation degree by matching the actual change with the expected change. This clearly reflects whether the actual operational changes of the motor under test conform to the changes expected of a healthy motor. It does not analyze solely based on the magnitude of a single data point, but rather analyzes the consistency of the trend and magnitude of changes, providing a more comprehensive reflection of the motor's operating status. The deviation degree calculated in this way accurately reflects the extent to which the motor under test deviates from healthy operating patterns, making anomaly detection more objective.

[0029] Preferably, the calculation of the normal index to be measured includes:

[0030] Obtain all target vectors of the motor under test within a preset time period, and locate the median value, denoted as the median vector to be tested; calculate the distance between the target vector to be tested and the median vector to be tested of the motor under test at time t within the preset time period, denoted as the deviation distance; add the square of the deviation distance to the square of the deviation of the motor under test at time t, and take the square root of the sum to obtain a total deviation to be tested; take the negative value of the total deviation to be tested, and calculate the exponential function with the natural constant as the base, thereby obtaining the normal index of the motor under test at time t.

[0031] Secondly, the present invention provides an Internet of Things-based motor data anomaly detection system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned Internet of Things-based motor data anomaly detection method is implemented.

[0032] By adopting the above technical solution, a computer program for detecting abnormal motor data based on the Internet of Things is generated and stored in a memory so that it can be loaded and executed by a processor. This allows for the creation of a terminal device based on the memory and processor, making it convenient to use.

[0033] The beneficial effects of this invention are as follows: Remote unmanned pumping stations are core infrastructure for ensuring water supply for residents, agricultural irrigation, and sewage treatment. The stable operation of their internal motor groups is directly related to residents' daily lives, agricultural production, and ecological environment governance. The IoT-based motor data anomaly detection method and system provided by this invention can achieve fully automated monitoring of the motor group throughout the entire process. It eliminates the need for frequent staff visits to remote pumping stations for inspections. The system can collect key motor data in real time, automatically analyze motor status, and identify early abnormal signals such as bearing wear and winding faults. This method not only avoids pumping station shutdowns due to escalating faults, reducing problems such as water outages for residents, delays in agricultural irrigation, and stagnation in sewage treatment, but also reduces maintenance costs and labor costs after motor failures. In the long term, this method and system can ensure the continuous and stable operation of pumping stations, providing strong support for people's livelihood, agricultural production, and ecological protection, and has significant practical value for maintaining normal social operation and promoting sustainable economic and social development. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating an IoT-based method for detecting abnormal motor data according to the present invention. Detailed Implementation

[0035] This invention discloses a method for detecting abnormal motor data based on the Internet of Things (IoT), referring to... Figure 1 This includes steps S1-S4:

[0036] S1: Synchronously acquire sensor data from different positions of the motor at a fixed frequency within a preset time period, arrange the data by time, and obtain multi-dimensional time-series data.

[0037] It should be noted that in municipal pumping stations in the industrial sector, there are numerous motors operating under complex and varied conditions, with significant differences in the normal operating parameter ranges of motors under different conditions. Analyzing motor data from different conditions together can lead to interference between normal and abnormal data, making it impossible to accurately determine whether a motor is faulty. For example, during the motor startup phase, current and vibration data are typically slightly higher than during stable operation. Analyzing startup data alongside stable operation data might mistakenly identify the normally high parameter values ​​during startup as abnormal, resulting in false alarms. Therefore, it is crucial to first acquire multi-dimensional motor data and align it over time. This invention selects temperature, current, and vibration data as core data sources because these three types of data can reflect the motor's operating status from different dimensions. Vibration data effectively reflects the operating status of rotating components such as motor bearings and rotors. If these components experience wear, imbalance, or other faults, changes in the vibration signal will lead to significant changes in the vibration data. Current data reflects the operating state of the motor stator windings; short circuits or open circuits in the windings will cause abnormal current. Temperature data indirectly reflects the motor's internal heat dissipation and overall operating load; abnormally high temperatures may indicate internal motor faults or excessive load. Setting the data acquisition frequency to 1kHz captures subtle parameter changes during motor operation without causing excessive data volume due to a higher acquisition frequency, thus avoiding increased data storage and processing pressure.

[0038] Specifically, a triaxial vibration sensor is installed on the bearing end cover of the motor, a current sensor is connected in series in the power supply line of the motor, and a temperature sensor is attached to the surface of the motor housing near the stator. Temperature data, current data, and vibration data from the sensors at different positions of the motor are acquired synchronously at 1kHz within a preset time period, and time-aligned. The data are then arranged in order according to the timestamps to form multidimensional time-series data, which includes temperature data sequences, current data sequences, and vibration data sequences.

[0039] Thus, multidimensional time-series data were obtained.

[0040] S2: The median values ​​of each dimension of the multidimensional time series data and the data of each dimension at the same time are denoted as the median data vector and the target data vector. Based on the relationship and difference between the median data vector and the target data vector, the initial covariance matrix is ​​obtained, the Mahalanobis distance sequence is calculated and obtained, and the abnormal boundary point is calculated according to the distance interval to obtain the normal motor group.

[0041] It should be noted that there are two problems when monitoring the motor groups of remote unattended pumping stations: First, the operating conditions of the motor groups are complex and changeable, such as fluctuations in water supply load at different times and motor start-stop switching; second, the multi-dimensional operating parameters of the motors, such as temperature, current and vibration, are inherently related, and traditional detection methods are prone to misjudging the motor operating status because they ignore the correlation between parameters or are not adapted to the dynamically changing operating conditions. This invention constructs a median data vector reflecting the normal operating benchmark of a motor group by calculating the median values ​​of temperature, current, and vibration data. Simultaneously, it integrates multi-dimensional actual data collected synchronously at various times into a target data vector, clearly defining the comparison object between the actual operating state of the motors and the normal benchmark of the group. Then, using the standard sample covariance matrix formula, it calculates the set of all target data vectors to analyze the inherent correlation between multi-dimensional parameters, avoiding limitations arising from judging based on a single parameter. Subsequently, based on this initial covariance matrix, it uses the Mahalanobis distance method to calculate the distance between all target data vectors and the median data vector in the multi-dimensional time series data, further solving the problem of inaccurate deviation calculation caused by parameter correlation. Finally, it calculates the Mahalanobis distance of all motors and arranges them in ascending order to form a Mahalanobis distance sequence, which is used to identify abnormal motors with significant deviations from the motor group.

[0042] Specifically, the median values ​​of temperature, current, and vibration are calculated for the temperature, current, and vibration data sequences in the multidimensional time-series data. These median values ​​are then combined to form a median data vector. Data with the same timestamp from the temperature, current, and vibration data sequences are combined and denoted as target data vectors. The initial covariance matrix is ​​obtained by calculating the covariance matrix for all data in the target time-series data set composed of all target data vectors using the standard sample covariance matrix formula. Based on the initial covariance matrix, the distance between all target data vectors and the median data vector in the multidimensional time-series data is calculated using the Mahalanobis distance method and denoted as the Mahalanobis distance. The Mahalanobis distances for all motors are calculated, and all obtained Mahalanobis distances are arranged from smallest to largest to obtain a Mahalanobis distance sequence.

[0043] It should be noted that, in order to automatically locate the core boundary that distinguishes normal and abnormal motors, it is necessary to calculate the difference between two adjacent values ​​in the Mahalanobis distance sequence, and then find the position corresponding to the maximum difference from all adjacent differences through the function of maximizing the independent variable. This position is the key gradient point in the Mahalanobis distance sequence where the distance changes most drastically, i.e. the abnormal boundary point. This process does not require manual setting of thresholds and is completed automatically by the system. It not only meets the detection requirements of unattended scenarios without human intervention, but also accurately locates the boundary between normal and abnormal motors.

[0044] Preferably, based on the magnitude of the difference between data points in the Mahalanobis distance sequence, the system automatically identifies the point with the largest distance gradient in the Mahalanobis distance sequence, which is recorded as the anomaly boundary point, including:

[0045] Anomaly boundaries satisfy the following expression:

[0046] ;

[0047] In the formula, It is the outlier boundary point in the Mahalanobis distance sequence, representing the key gradient point in the Mahalanobis distance sequence where the distance changes most drastically; , It is the th in the Mahalanobis distance sequence , A number; It represents the difference between two adjacent values ​​in the Mahalanobis distance sequence, reflecting the amount of change in Mahalanobis distance at adjacent positions; It is a function that maximizes the independent variable, and it can maximize all possible values. Found The most dramatic changes .

[0048] Preferably, the abnormal boundary point is obtained in the order i of the Mahalanobis distance sequence, the points after the i-th gradient point in the Mahalanobis distance sequence are removed, the remaining gradient points are recorded as the standard point set, and the motors corresponding to all points in the standard point set are recorded as the normal motor group.

[0049] At this point, the normal motor group has been identified.

[0050] S3: Obtain the normal median vector and normal target vector of a single motor in the normal motor group; calculate the health response characteristics based on the difference between the change range of the normal target vector of a single motor in a unit time and the deviation of the normal target vector at the previous moment; calculate the normal expected vector based on the influence of the health response characteristics on the correction process of the normal target vector.

[0051] It should be noted that after obtaining the normal motor group, it is necessary to further explore the operating patterns of the healthy motors. By obtaining the normal target vector of an individual motor in the normal motor group (i.e., the real-time operating state and the normal median vector, i.e., its own operating baseline), the difference between the normal target vectors at adjacent time points is calculated to obtain the normal instantaneous vector, which reflects the actual state change per unit time. The difference between the normal median vector and the normal target vector at the previous time point is calculated to obtain the historical driving vector, which reflects the theoretical change required to regress to the baseline. The two are compared to obtain the health response characteristics, which are used to describe the unique response speed and correction of the healthy motor when facing deviations in different dimensions. This provides the core basis for subsequently establishing a state prediction model for healthy motors and accurately judging whether the motor under test is abnormal.

[0052] Preferably, the target data vector and the median data vector within a preset time period corresponding to any moment of a single motor in a normal motor group are obtained, denoted as the normal target vector and the normal median vector, respectively; the historical drive vector is calculated based on the difference between the normal target vector and the normal median vector; the normal instantaneous vector is calculated based on the difference between two normal target vectors at adjacent times; and the health response characteristics are calculated based on the difference between the historical drive vector and the normal instantaneous vector, including:

[0053] It should be noted that healthy motors exhibit significant differences in their abnormal recovery patterns across the dimensions of temperature, current, and vibration. Traditional testing methods using only a single indicator cannot reflect these dimensional characteristics. The formula, by describing the correction percentage of each dimension's data, indicates the need to explore the multi-dimensional operating patterns of healthy motors.

[0054] The health response characteristics satisfy the following expression:

[0055] ;

[0056] In the formula, This represents the health response characteristics of a single motor at time t in the multidimensional time-series data of a normal multidimensional dataset. , This represents the normal target vector at time t and time t-1 in the multidimensional time series data of a single motor in a normal multidimensional dataset; The normal median vector represents the multidimensional time-series data of a single motor in a normal multidimensional dataset.

[0057] In the formula, It is represented as a multi-dimensional vector, with each component corresponding to the response characteristics in physical dimensions such as temperature, current, and vibration, revealing the inherent response speed and behavior of a healthy motor in different dimensions; This represents a normal instantaneous vector, used to reflect the changes in the motor state of a normal motor within a unit of time. The historical driving vector represents the amount of change required to cause the motor state to revert from its current position to the normal median vector at time t-1 in the multidimensional time series data of a single motor in a normal multidimensional dataset. Let represent the percentage of the actual normal instantaneous vector of a single motor at time t-1 in the multidimensional time series data of a normal multidimensional dataset, relative to the theoretically required correction due to deviation from the normal median vector. A more complete description can be given of how a normal motor should self-correct in its different physical dimensions when faced with deviations, each with its own unique characteristics.

[0058] It should be noted that due to the dynamic fluctuations in the operating conditions of motors in remote, unattended pumping stations, water supply loads such as morning and evening peak water usage, irrigation cycles, and changes in ambient temperature will cause the parameters of healthy motors, such as current and temperature, to exhibit dynamic ranges. Using the current state at a specific moment as a fixed benchmark can easily lead to misjudgments. Simply knowing the current state of a healthy motor is insufficient to determine whether a motor under test is abnormal; a state prediction model for the healthy motor is needed to clarify its expected operating state. Therefore, based on the normal target vector and healthy response characteristics of a normal motor at time t, the difference between the normal target vector and the normal median vector at that moment is first calculated to obtain the instantaneous regression driving vector. Then, the healthy response characteristics are multiplied by this correction amount to obtain the expected change that the healthy motor actually needs to complete in the next unit of time. Finally, the current normal target vector and the expected change amount are added to obtain the normal expected vector at time t+1. This process constructs a state evolution model for the healthy motor, clarifying the expected operating state of the healthy motor at different times, providing a reference for subsequently comparing the actual state and expected state of the motor under test and determining whether it is abnormal.

[0059] Preferably, based on the normal target vector and healthy response characteristics at time t in the multidimensional time series data of a single motor in the normal multidimensional dataset, the normal expected vector at time t+1 in the multidimensional time series data of a single motor in the normal multidimensional dataset is predicted, denoted as the normal expected vector, which includes:

[0060] It should be noted that pump station operating conditions are highly variable, such as fluctuations in water supply load causing dynamic changes in motor parameters. The current state of a healthy motor alone cannot provide a reference for the normal operating state of the motor under test at the next moment; this step involves... The actual effect of dynamically adjusting the theoretical correction amount is to ensure... It can adapt to the health standards under current working conditions.

[0061] The normal expected vector satisfies the following expression:

[0062] ;

[0063] In the formula, , , This represents the normal expected vector, the normal target vector, and the health response characteristics at time t+1 in the multidimensional time series data of a single motor in a normal multidimensional dataset. This represents the normal median vector in the multidimensional time-series data of a single motor within a normal multidimensional dataset.

[0064] In the formula, This represents the difference between the normal target vector and the normal median vector of the motor at time t, denoted as the instantaneous regression drive vector, which is used to describe the theoretical correction amount required to restore the motor to normal. It applies the health response characteristics at time t to the instantaneous regression driving vector, representing the actual correction action that a relatively normal motor needs to complete in the next time unit when faced with the current theoretical correction amount, combined with its own actual situation, and is denoted as the expected change amount. The normal expected vector at time t+1 is represented by adding the normal target vector of a single normal motor at time t to the expected change that needs to be actually completed in the next time unit.

[0065] S4: Obtain the target vector and expected vector of any motor under test; calculate the deviation degree based on the consistency between the actual change of the target vector and the theoretical change that needs to be corrected to the expected vector; calculate the normality index based on the impact of the deviation degree on the target vector and the difference between the median vector and the target vector; continuously monitor and issue early warnings for anomalies.

[0066] It is important to note that the core of determining whether a motor under test is abnormal lies in whether its actual operating state changes conform to the state evolution law of a healthy motor. Based on the normal expected vector model established in the previous section, the target vector of the motor under test at time t and time t+1 is obtained, and the difference between the two is calculated to obtain the actual change, which reflects the true state change of the motor under test per unit time. At the same time, based on the target vector of the motor under test at time t and the expected vector at time t+1, the difference between the two is calculated to obtain the expected change, which reflects the state change that a healthy motor should have under the same conditions. By calculating the cosine similarity between the actual change and the expected change, the consistency of their direction and amplitude can be determined. The closer the similarity is to 1, the more the state change of the motor under test conforms to the healthy law. Subtracting the similarity from 1 yields the deviation. The closer the deviation is to 1, the greater the deviation between the actual change of the motor under test and the healthy law, providing an evaluation standard for subsequent comprehensive judgment of the degree of motor abnormality.

[0067] Specifically, one motor is randomly selected from all the motors under test, and its target data vector is obtained, denoted as the target vector under test; the normal expected vector of the target vector under test at any time is calculated, denoted as the expected vector under test; based on the difference between the expected vector under test and the target vector under test, the deviation under test is calculated, including:

[0068] It should be noted that the temperature, current, and vibration data of the motor under test vary significantly and are strongly correlated; for example, an increase in current is accompanied by an increase in temperature. Traditional testing methods, which only compare changes in a single parameter, are prone to misjudgment. This invention first determines whether the actual direction of change is consistent with the expected direction of change of a healthy motor using cosine similarity, and then converts it into an intuitive... This allows for automatic analysis of anomalies and meets the requirements for fully automated parameter comparison under unattended operation. The expression for the deviation under test not only reflects the difference between the changes in the state of the motor under test and its health patterns, but also reduces the differences and correlation interference between parameter magnitudes through vector dot product and modulus, avoiding the limitation of parameter magnitude on the accuracy of traditional distance calculation.

[0069] The deviation to be measured satisfies the following expression:

[0070] ;

[0071] In the formula, This represents the measured deviation of the motor under test at time t within a preset time period; , Let represent the target vector to be measured at time t and time t+1; This represents the expected vector to be measured at time t+1. This is the modulus symbol.

[0072] In the formula, This represents the difference between the target vector of the motor under test at time t and time t, which describes the actual change in the target vector of the motor under test per unit time, and is denoted as the actual change. It represents the difference between the expected vector of the motor under test at time t+1 and the target vector of the motor under test at time t. It is used to describe the amount of vector change that the motor under test needs to achieve to satisfy the expected vector of the motor under test, and is denoted as the expected change. express The vector magnitude; express The vector magnitude; The cosine similarity between the actual change and the expected change is represented. The closer the value is to 1, the more symmetrical the two vectors are, and the more normal the motor condition is. The closer the value is to 1, the greater the deviation and the more abnormal the motor is.

[0073] It should be noted that although some motors may exhibit healthy behavior in their state changes, their overall operating status may deviate from their median level over a long period, indicating potential anomalies. Therefore, judging the motor's condition solely based on the measured deviation is insufficient. It is also necessary to consider the deviation of the motor's operating status from its own baseline. Thus, the median value of all target vectors within a preset time period is first obtained as the measured median vector. The distance between the target vector and the measured median vector at time t is calculated to obtain the deviation distance. The square of the deviation distance is then added to the square of the measured deviation at that time, and the square root is taken to obtain the total measured deviation, which comprehensively evaluates the deviation from the baseline and the deviation from the healthy behavior. Finally, the total deviation is negatively evaluated, and an exponential function with the natural constant as the base is calculated to obtain the measured normality index. The smaller the exponent, the higher the overall degree of anomaly of the motor under test.

[0074] Specifically, all target vectors of the motor under test within a preset time period are acquired, and the median value is located and denoted as the median vector. Based on the influence of the deviation on the target vector of the motor under test, and the difference between the median vector and the target vector, the normality index of the motor under test is calculated and denoted as the normality index, including:

[0075] It should be noted that when conducting a comprehensive and intuitive analysis of motor status, some motors may exhibit status changes that conform to healthy patterns, but may still have potential anomalies due to long-term deviations from their own operating benchmarks. This invention addresses the pain points of untimely warnings and high misjudgment rates in traditional detection by integrating deviations from two dimensions and then converting them into a test normal index. The expression for the test normal index not only integrates multi-dimensional abnormal information but also amplifies the distinction between abnormal and normal conditions through an exponential function, avoiding the judgment confusion caused by the superposition of multiple indicators in traditional methods.

[0076] The normal index to be tested satisfies the following expression:

[0077] ;

[0078] In the formula, This represents the normal operation index of the motor under test at time t within a preset time period; This represents the target vector of the motor under test at time t within a preset time period; Represents the midpoint vector to be measured; This represents the measured deviation of the motor under test at time t within a preset time period; This is the modulus symbol.

[0079] In the formula, The distance between the target vector of the motor under test and the midpoint vector under test at time t is denoted as the deviation distance. This represents the square of the measured deviation of the motor under test at time t; This means that the square root of the sum of the square of the distance between the target vector and the median vector of the motor under test at time t and the square of the deviation at that time can be understood from a geometric perspective as the total deviation of the motor under test, which combines the spatial distance between the target vector and the median vector and the deviation. This indicates that the normality of the motor under test at time t is comprehensively quantified by using vector magnitude, measured deviation, and exponential function, providing a more comprehensive indicator for motor condition monitoring. This is denoted as the measured normality index. The smaller the value, the smaller the normal index of the motor under test, and the more abnormal the motor under test.

[0080] It's important to note that the core requirement for remote, unmanned pumping stations is to achieve fully automated, real-time response to motor anomalies, eliminating the need for manual intervention in threshold setting and judgment. Based on the normal index obtained in the preceding section, and combined with the normal index distribution range generated from historical monitoring data of normal motor groups, the normal index of the motor under test is continuously monitored using the real-time and continuous capabilities of the Internet of Things (IoT). When the normal index of the motor under test falls outside the normal index distribution range, an anomaly can be determined. The IoT system will push the anomaly determination result to the terminal in real time, automatically triggering fault warnings and subsequent diagnostic processes. Ultimately, this achieves full-process intelligentization from data collection, benchmark establishment, anomaly judgment to warning response, which to some extent solves the problems of difficult monitoring, untimely response, and high misjudgment rate of motor groups in remote, unmanned scenarios.

[0081] Preferably, based on the real-time and continuous nature of the Internet of Things, the motor under test... Continuous monitoring will be conducted; utilizing the normal index distribution range generated from historical monitoring of normal motor groups, when the motor under test... When the motor falls outside the normal index distribution range, it is determined that the motor is abnormal. The Internet of Things system pushes the abnormality determination result to the terminal in real time, triggering fault warning and subsequent diagnosis process, and finally realizing fully automatic intelligent detection and response to motor abnormalities without human threshold intervention.

[0082] This completes the accurate detection of abnormal motor data.

[0083] This invention also discloses an IoT-based motor data anomaly detection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an IoT-based motor data anomaly detection method according to this invention.

[0084] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0085] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for detecting motor data anomaly based on Internet of Things, characterized in that, The application relates to a method for monitoring and early warning of motor faults. Synchronously acquiring sensor data at different positions of the motor at a fixed frequency within a preset time, arranging the data according to time, and obtaining multi-dimensional time sequence data; The median value of each dimension data in the multi-dimensional time sequence data and the data at the same time are recorded as a median data vector and a target data vector; an initial covariance matrix is obtained according to the relationship and difference between the median data vector and the target data vector; a Mahalanobis distance sequence is calculated and obtained; an abnormal boundary point is calculated according to the distance interval size; and a normal motor group is obtained. The normal median vector and the normal target vector of a single motor in the normal motor group are obtained; a health response feature is calculated according to the difference between the change amplitude of the normal target vector of the single motor within a unit time and the deviation degree of the normal target vector at a previous time; the normal expected vector is calculated according to the influence of the health response feature on the correction process of the normal target vector. ； In the formula, represents the health response feature of the t-th time in the multi-dimensional time series data of a single motor in the normal multi-dimensional data set; , represents the normal target vector of the t-th time and the t-1-th time in the multi-dimensional time series data of a single motor in the normal multi-dimensional data set; represents the normal median vector of the multi-dimensional time series data of a single motor in the normal multi-dimensional data set; The target vector and the expected vector of an arbitrary motor to be measured are obtained; the measured deviation degree is calculated according to the consistency between the actual change of the target vector to be measured and the theoretical change required to correct to the expected vector to be measured; the measured normal index is calculated according to the influence of the measured deviation degree on the target vector to be measured and the difference between the measured median vector and the target vector to be measured; and continuous monitoring and abnormal early warning are carried out. The multi-dimensional time sequence data is obtained by: 2.The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that, A three-axis vibration sensor is installed at the bearing end cover of the motor, a current sensor is connected in series in the power supply circuit of the motor, and a temperature sensor is attached to the surface of the motor shell near the stator; temperature data, current data and vibration data on the sensors corresponding to different positions of the motor within a preset time are synchronously acquired at a fixed frequency; the data are time-aligned; the data are arranged in sequence according to the time stamps of the data; and multi-dimensional time sequence data is formed, wherein the multi-dimensional time sequence data comprises a temperature data sequence, a current data sequence and a vibration data sequence. The median value of each dimension data in the multi-dimensional time sequence data and the data at the same time are recorded as a median data vector and a target data vector, which comprises: 3.The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that, The temperature median value, the current median value and the vibration median value of the temperature data sequence, the current data sequence and the vibration data sequence in the multi-dimensional time sequence data are calculated respectively; the temperature median value, the current median value and the vibration median value are combined to form a median data vector; and the data with the same time stamp in the temperature data sequence, the current data sequence and the vibration data sequence are combined to form a target data vector. The initial covariance matrix is obtained by calculating all the data in the target time sequence data set formed by all the target data vectors by using a standard sample covariance matrix formula; the distance between all the target data vectors and the median data vector in the multi-dimensional time sequence data is calculated by using the Mahalanobis distance method based on the initial covariance matrix, and is recorded as a Mahalanobis distance; the Mahalanobis distances of all the motors are calculated, and the obtained Mahalanobis distances are arranged from small to large to obtain a Mahalanobis distance sequence. 4.The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that, The abnormal boundary point is calculated, and the normal motor group is obtained, which comprises: ​ 5. The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that, ​ Difference is performed on all adjacent two Mahalanobis distance values in the Mahalanobis distance sequence to obtain a plurality of difference values; a maximum difference value is determined in the plurality of difference values, and a position point corresponding to the maximum difference value and located in the Mahalanobis distance sequence is determined as an abnormal boundary point; an order i of the abnormal boundary point in the Mahalanobis distance sequence is obtained, points after the i-th gradient point in the Mahalanobis distance sequence are removed, and the remaining gradient points are recorded as a standard point set; and the motors corresponding to all points in the standard point set are recorded as a normal motor group. 6.The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that, The calculating the normal expected vector comprises: According to the normal target vector at the t-th moment and the health response feature in the multi-dimensional time sequence data of the single motor in the normal multi-dimensional data set, a difference value between the normal target vector at the t-th moment and the normal median vector is calculated, and is recorded as an instantaneous regression driving vector; the health response feature of the single motor at the t-th moment is multiplied by the instantaneous regression driving vector to obtain an expected change amount; and the normal target vector of the single motor at the t-th moment is added to the expected change amount to obtain the normal expected vector of the single motor at the (t+1)-th moment.

7. The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that, The to-be-measured deviation degree satisfies an expression: ； In the formula, represents the deviation to be measured of the motor to be measured at the tth moment within the preset time; , represents the target vector to be measured at the tth moment and the t+1th moment; represents the expected vector to be measured at the t+1th moment; is a modulus length symbol. 8.The motor data anomaly detection method based on the Internet of Things according to claim 1, characterized in that The calculating the to-be-measured normal index comprises: All to-be-measured target vectors of the to-be-measured motor within a preset time are obtained, and a median value is located, and is recorded as a to-be-measured median vector; a distance between the to-be-measured target vector of the to-be-measured motor at the t-th moment within the preset time and the to-be-measured median vector is calculated, and is recorded as a deviation distance; a square of the deviation distance is added to a square of the to-be-measured deviation degree of the to-be-measured motor at the t-th moment, and a square root of a result of the addition is taken to obtain a to-be-measured total deviation degree; a negative value of the to-be-measured total deviation degree is taken, and an exponential function with a natural constant as a base is calculated to obtain the to-be-measured normal index of the to-be-measured motor at the t-th moment.

9. An Internet of Things-based motor data anomaly detection system, characterized in that, Comprise: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, a kind of motor data anomaly detection method based on Internet of Things according to any one of claims 1-8 is realized.

Images