A Data Analysis-Based Method and System for Detecting the Operating Status of a Dehydrator

Abstract

This invention relates to the field of electronic data processing technology, and discloses a method and system for detecting the operating state of a dehydrator based on data analysis. The method includes acquiring multidimensional vibration signals from the dehydrator during operation and extracting features to obtain a set of feature vectors; determining the neighborhood radius of density clustering based on the inverse matrix of the covariance matrix of the k-nearest neighbor set of each feature vector; calculating the local structural entropy and performing density clustering on the feature vectors within the current time window to obtain the current cluster set; constructing a bipartite graph between the current cluster set and the cluster set of the previous time window; identifying persistent clusters, newly formed clusters, and extinct clusters; establishing a Kalman filter model for the time series of the system state transition index; and determining that the operating state of the dehydrator has changed when the cumulative sum of the predicted innovation sequence and the control chart statistics exceed a preset control limit for N consecutive sampling points. This invention can suppress noise interference, identify persistent trends caused by actual state changes, and improve the reliability of state change detection.

Description

Technical Field

[0001] This invention relates to the field of electronic data processing technology, and specifically to a method and system for detecting the working status of a dehydrator based on data analysis. Background Technology

[0002] During actual operation, dewatering machines are prone to abnormal deviations in their working state due to factors such as fluctuations in material properties, mechanical wear of core components, or improper human operation. If these changes are not accurately identified and addressed in a timely manner, it will not only directly lead to a decrease in the dewatering efficiency of the dewatering machine and excessive moisture content in the filter cake, but may also indirectly affect the stability of subsequent production processes and increase overall energy consumption costs.

[0003] Currently, vibration signal analysis is the mainstream technique for monitoring the operating status of rotating machinery. Traditional methods typically involve first collecting vibration signals from one or more measuring points during equipment operation, extracting key features in the time domain, frequency domain, or time-frequency domain, and then determining whether the equipment is in a normal state through threshold comparison, expert systems, or traditional machine learning classification algorithms. Traditional machine learning classification algorithms include support vector machines and neural networks. However, in actual production, fault samples are often difficult to obtain, and traditional methods are ill-suited to the uneven density, blurred boundaries, and diverse local structures of feature data under different loads and speeds in dehydrators.

[0004] The change in the operating state of a dehydrator is not instantaneous, but a continuous, cumulative process, specifically manifested as a gradual change in the data distribution pattern in the feature space. If static cluster analysis is performed only on data from a single time window, it is impossible to capture the continuity and trend of data distribution changes, making it difficult to effectively distinguish between fluctuations in normal operating conditions and early signs of abnormal states. Furthermore, it is also impossible to differentiate between trend changes caused by changes in the system's internal state and transient fluctuations caused by random noise, resulting in low accuracy in identifying changes in the dehydrator's operating state. Summary of the Invention

[0005] This invention provides a data analysis-based method and system for detecting the working status of a dehydrator to solve the problem of low accuracy in identifying changes in the working status of a dehydrator in the prior art.

[0006] In a first aspect, the data analysis-based method for detecting the operating status of a dehydrator according to the present invention includes the following steps:

[0007] The multidimensional vibration signal of the dehydrator during operation is acquired, and the vibration signal is divided into continuous time windows with overlap. The features of the data in each time window are extracted to obtain a set of feature vectors.

[0008] The distance is determined by the inverse of the covariance matrix of the k nearest neighbor set of each feature vector, and the neighborhood radius of density clustering is determined based on the distance. The local structure entropy is obtained by the Shannon entropy of the normalized eigenvalues ​​of the covariance matrix of the k nearest neighbor set. Feature vectors with local structure entropy lower than a first threshold and the number of data points in their neighborhood higher than a minimum threshold are selected as core points. Density clustering is performed on the feature vectors in the current time window using the core points and the neighborhood radius to obtain the current cluster set.

[0009] Construct a bipartite graph between the current cluster set and the cluster set of the previous time window. The weights of the edges in the bipartite graph are determined by the weighted sum of the intersection-union ratio of the point sets of the two clusters, the distance between the point sets, and the distance between the centroids of the clusters. By solving the minimum weight matching problem of the bipartite graph, establish the mapping relationship between the clusters and identify persistent clusters, newly formed clusters, and extinct clusters.

[0010] Calculate the system state transition index and establish a Kalman filter model of the system state transition index time series. When the cumulative sum of the predicted information sequence of the Kalman filter model and the control chart statistics exceed the preset control limits for N consecutive sampling points, it is determined that the working state of the dehydrator has changed.

[0011] Preferably, the calculation of the system state transition index includes: the system state transition index is calculated using the following formula, ;

[0012] in, This represents the system state transition index for the current time window. λ is the system state transition exponent for the previous time window, and λ is the forgetting factor. This represents the energy increment for the current time window.

[0013] The energy increment is obtained by weighting the matching cost of all persistent clusters, the intra-cluster dispersion of all newly formed clusters, and the historical duration of all extinct clusters.

[0014] Preferably, the step of determining the distance based on the inverse of the covariance matrix of the k nearest neighbor set of each feature vector, and determining the neighborhood radius of density clustering based on the distance, includes: for feature vectors Calculate its k-nearest neighbor set covariance matrix Any two eigenvectors and The distance between them is calculated using the following formula.

[0015] ;

[0016] in, For feature vectors and The distance between, for and The transpose of the difference vector, for The covariance matrix, for The inverse matrix; for The inverse of the covariance matrix of the set of k nearest neighbors;

[0017] Calculate the distance from each point in the dataset to its k-th nearest neighbor, arrange all distance values ​​in ascending order and plot the k-dist curve, and select the distance value corresponding to the inflection point of the curve as the neighborhood radius of density clustering.

[0018] Preferably, obtaining the local structural entropy based on the Shannon entropy of the normalized eigenvalues ​​of the covariance matrix of the k-nearest neighbor set includes: calculating the eigenvectors. The covariance matrix of the k-nearest neighbor set d eigenvalues Normalize the eigenvalues ​​to obtain P1, P2, and P3. P d ; Calculate using the following formula Local structural entropy ,

[0019] ;in, This is the normalized value of the j-th eigenvalue.

[0020] Preferably, the energy increment is obtained by a weighted sum of the matching costs of all persistent clusters, the intra-cluster dispersion of all newly formed clusters, and the historical duration of all extinct clusters, including: summing the matching costs of all persistent clusters in the minimum weight matching of the bipartite graph. Calculate the average Euclidean distance from all points within each new cluster to the centroid. This average Euclidean distance is used as the intra-cluster scatter. Summing the intra-cluster scatter of all new clusters yields the result. ; Obtain the intra-cluster scatter of each dying cluster in the previous time window, and sum the intra-cluster scatter of all dying clusters to obtain the result. Energy increment It is calculated using the following formula.

[0021] ;

[0022] in, , , These are the preset weighting coefficients.

[0023] Preferably, determining that the dehydrator's operating state has changed when the cumulative sum of the control chart statistics of the predicted innovation sequence from the Kalman filter model exceeds the preset control limit for N consecutive sampling points includes: obtaining the predicted innovation sequence and its standard deviation using the Kalman filter model. The cumulative control chart statistic is calculated using the following formula.

[0024] ;

[0025] in, Let be the cumulative sum and control chart statistic at time t, and let k be the reference value. , for Accumulated data over time and control chart statistics. For the predicted information at time t;

[0026] When cumulative sum control chart statistics When five consecutive sampling points exceed the control limit H, the dehydrator's operating state is determined to have changed. The control limit H is... .

[0027] Preferably, acquiring the multidimensional vibration signal during the operation of the dehydrator includes: installing a triaxial accelerometer on the bearing housing of the dehydrator to collect the vibration acceleration signals of the dehydrator in the X, Y, and Z directions.

[0028] Preferably, the extraction of features of data within each time window includes: calculating the time-domain features and frequency-domain features of the vibration acceleration signals in the X, Y, and Z directions within each time window, respectively. The time-domain features include root mean square, kurtosis, margin, and peak-to-peak value, while the frequency-domain features include spectral entropy, spectral kurtosis, and centroid frequency.

[0029] Preferably, the overlap rate is 50%.

[0030] Secondly, the data analysis-based dehydrator operating status detection system of the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the above-mentioned data analysis-based dehydrator operating status detection method.

[0031] The beneficial effects of this invention are as follows: First, by defining distance and local structural entropy based on the covariance matrix of the k-nearest neighbor set, this invention improves the density clustering process, effectively handling the uneven density and ambiguous boundary characteristics of data generated by dehydrators under complex operating conditions, thus improving the accuracy of clustering. Second, by constructing a bipartite graph between time-window cluster sets and solving for minimum weight matching, the temporal relationship of data cluster changes is established, enabling quantitative identification of cluster maintenance, generation, and extinction processes. The system state transition index integrates the matching cost of persistent clusters, the dispersion of newly generated clusters, and the survival information of extinct clusters, transforming complex cluster changes into an indicator highly sensitive to changes in system stability. Third, this invention can suppress the interference of random noise and identify weak persistent trends caused by real state changes, thereby improving the timeliness and reliability of state change detection. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the data analysis-based method for detecting the working status of a dehydrator, as provided in an embodiment of the present invention. Detailed Implementation

[0033] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0034] like Figure 1 As shown, an embodiment of the data analysis-based dehydrator operating status detection method provided by the present invention includes the following steps:

[0035] S1. Acquire the multidimensional vibration signal during the operation of the dehydrator, divide the vibration signal into continuous time windows with overlap, extract the features of the data in each time window, and obtain a set of feature vectors.

[0036] Specifically, triaxial accelerometers are installed at key locations such as the bearing housing of the dewatering machine to collect vibration acceleration signals in the X, Y, and Z directions. Preferably, the time window length is 2048 sampling points with an overlap rate of 50%, meaning each new window overlaps with the previous window by 1024 sampling points. For the vibration acceleration signals in the X, Y, and Z directions within each time window, time-domain features such as mean, root mean square, variance, kurtosis, kurtosis, peak-to-peak value, and margin are calculated, along with frequency-domain features such as spectral centroid, centroid frequency, spectral entropy, spectral kurtosis, spectral kurtosis, frequency standard deviation, and spectral flatness. All features are then combined into a feature vector. By simultaneously extracting features in the time and frequency domains, the state information in the signal can be understood from different perspectives. For example, time-domain features reflect vibration energy and impact, while frequency-domain features represent the distribution pattern and periodicity of vibration energy at different frequencies.

[0037] For example, taking a time window of 1 second as an example, assuming the vibration signal of a dehydrator along the X-axis is collected at a sampling frequency of 2048 Hz, a time series containing 2048 data points is obtained. In the time domain, the mean of the 2048 points is 0.01, indicating that the DC component of the vibration is very small; the root mean square value is 0.5, reflecting the magnitude of the vibration energy; the kurtosis value is 4.2, and a value greater than 3 indicates the presence of an impulsive component in the signal. In the frequency domain, a fast Fourier transform is performed on the signal to obtain the spectrum, and the spectral centroid is calculated to be 150 Hz, indicating that the vibration energy is mainly concentrated around 150 Hz; the spectral entropy is 0.7, representing the complexity of the spectrum; and the spectral flatness is 0.2, indicating the presence of obvious harmonic peaks in the spectrum rather than uniformly distributed white noise. Combining the feature values ​​extracted from the vibration signals in the X, Y, and Z axes in the time and frequency domains yields a multidimensional feature vector. Repeat this process for all data segments collected within the time window to obtain a set of feature vectors, where each vector represents an instantaneous operating state of the device at that moment.

[0038] S2, determine the distance based on the inverse matrix of the covariance matrix of the k nearest neighbor set of each feature vector, and determine the neighborhood radius of density clustering based on the distance; obtain the local structure entropy based on the Shannon entropy of the normalized eigenvalues ​​of the covariance matrix of the k nearest neighbor set, and take the feature vectors whose local structure entropy is lower than the first threshold and whose number of data points in the neighborhood is higher than the minimum number threshold as core points; use the core points and neighborhood radius to perform density clustering on the feature vectors in the current time window to obtain the current cluster set.

[0039] S3. Construct a bipartite graph between the current cluster set and the cluster set of the previous time window. The weights of the edges in the bipartite graph are determined by the weighted sum of the intersection-union ratio of the point sets of the two clusters, the distance between the point sets, and the distance between the centroids of the clusters. By solving the minimum weight matching problem of the bipartite graph, establish the mapping relationship between the clusters and identify the persistent clusters, new clusters, and extinct clusters.

[0040] Specifically, the set of nodes on the left side of the bipartite graph is C. T-1 For all clusters, the right-hand node set is C. T All clusters; connect C T-1 Cluster i and C T Weights of edges in cluster j Calculated using the following formula:

[0041] ;

[0042] in, The ratio of the intersection to the union of the eigenvector sets contained in the two clusters. The Hausdorff distance between two point sets. Let be the Mahalanobis distance between the centroids of the two clusters, and the weighting coefficients a, b, and c are preferably 0.5, 0.3, and 0.2, respectively.

[0043] The Hungarian algorithm is used to find the minimum weight matching in a bipartite graph; if cluster i and cluster j successfully match, they form a persistent cluster; in C T Clusters that do not find a match in C are newly formed clusters; T-1 Clusters that do not find a match are considered extinct clusters.

[0044] S4. Calculate the system state transition index and establish a Kalman filter model of the system state transition index time series. When the cumulative sum of the predicted information sequence of the Kalman filter model and the control chart statistics exceed the preset control limits for N consecutive sampling points, it is determined that the working state of the dehydrator has changed.

[0045] In an optional embodiment, calculating the system state transition index includes: the system state transition index is calculated using the following formula, ;

[0046] in, This represents the system state transition index for the current time window. λ is the system state transition exponent for the previous time window, and λ is the forgetting factor. This represents the energy increment for the current time window.

[0047] The energy increment is obtained by weighting the matching cost of all persistent clusters, the intra-cluster dispersion of all newly formed clusters, and the historical duration of all extinct clusters.

[0048] In an optional embodiment, determining the distance based on the inverse of the covariance matrix of the k nearest neighbor set of each feature vector, and determining the neighborhood radius of density clustering based on the distance, includes: for feature vectors Calculate its k-nearest neighbor set covariance matrix Any two eigenvectors and The distance between them is calculated using the following formula.

[0049] ;

[0050] in, For feature vectors and The distance between, for and The transpose of the difference vector, for The covariance matrix, for The inverse matrix; for The inverse of the covariance matrix of the set of k nearest neighbors;

[0051] Calculate the distance from each point in the dataset to its k-th nearest neighbor, arrange all distance values ​​in ascending order and plot the k-dist curve, and select the distance value corresponding to the inflection point of the curve as the neighborhood radius of density clustering.

[0052] Distance can perceive the local distribution shape of data. For example, when data points are densely arranged along a certain direction, they are closer along that direction, while they are farther apart in sparse regions perpendicular to that direction, thus reflecting the true proximity relationship between data points. Suppose that in a two-dimensional feature space, data points are mainly distributed along a curve. For a point on the curve... Select the 20 nearest points as the k-nearest neighbor set, and calculate the covariance matrix of these 20 points. The covariance matrix reflects that the variance of the data is larger along the tangent direction of the curve and smaller along the normal direction. In calculating... With another point When calculating the distance, the inverse of the covariance matrix is ​​used to weight the displacements in different directions. If the displacement direction is consistent with the main direction of the data distribution, the weight is small; if the displacement direction is perpendicular to the main direction of the data distribution, the weight is large.

[0053] When determining the neighborhood radius for density clustering, first calculate the distance to its 20th nearest neighbor for each point in the dataset. Assuming there are 1000 points in the dataset, this yields 1000 such distance values. These 1000 distance values ​​are then sorted in ascending order and plotted as a k-dist curve. The horizontal axis of the curve represents the index of the data point, and the vertical axis represents the corresponding distance value. Typically, the curve rises gently at the beginning, representing the k-neighbor distance of points in the cluster core region. Then, a sudden inflection point appears, with a sharply increasing slope, indicating the beginning of encountering noisy points or sparse regions. The distance value corresponding to this inflection point is the neighborhood radius of the density cluster.

[0054] In an optional embodiment, obtaining the local structure entropy based on the Shannon entropy of the normalized eigenvalues ​​of the covariance matrix of the k-nearest neighbor set includes: calculating the eigenvectors. The covariance matrix of the k-nearest neighbor set d eigenvalues Normalize the eigenvalues ​​to obtain P1, P2, and P3. P d ; Calculate using the following formula Local structural entropy,

[0055] ;in, This is the normalized value of the j-th eigenvalue.

[0056] The eigenvalues ​​of the covariance matrix represent the dispersion or variance of the data along the principal component directions. If the data is locally linear or planar, most of the variance will be concentrated in one or two principal directions, resulting in one or two large eigenvalues ​​while the rest are small, leading to a low calculated entropy. Conversely, if the data is locally clumped or spherically uniformly distributed, the variances in all directions are very close, and the eigenvalues ​​are of similar magnitude, resulting in a high calculated entropy. For example, consider data points in a three-dimensional feature space. For example, find its k-nearest neighbor set, say k is 30, and calculate the k-nearest neighbor set of these 30 points. Covariance matrix. Through eigenvalue decomposition, we obtain three eigenvalues, assumed to be... It equals 8.5. It equals 1.0. The value equals 0.5. This indicates that the variability of the data in the first principal direction is much greater than in the other two directions, and the local data structure may approximate a line. To calculate the entropy, the eigenvalues ​​are first normalized; their sum is 10, and the normalized value is... It equals 0.85. It equals 0.1. It equals 0.05. Then, the local structural entropy is calculated using the Shannon entropy formula.

[0057] In an optional embodiment, the energy increment is obtained by a weighted sum of the matching costs of all persistent clusters, the intra-cluster dispersion of all newly formed clusters, and the historical duration of all extinct clusters, including:

[0058] The summation of the matching costs of all persistent clusters in the minimum weight matching of the bipartite graph is obtained. ;

[0059] Calculate the average Euclidean distance from all points within each new cluster to the centroid. This average Euclidean distance is used as the intra-cluster scatter. Summate the intra-cluster scatter of all new clusters to obtain the result. ;

[0060] Obtain the intra-cluster scatter of each dying cluster in the previous time window, and sum the intra-cluster scatter of all dying clusters to obtain the result. ;

[0061] Energy increment It is calculated using the following formula.

[0062] ;

[0063] in, , , These are the preset weighting coefficients.

[0064] The evolution of states is decomposed into three basic events: state persistence, emergence, and extinction, and the impact of each event on overall stability is evaluated. The matching cost of a persisting cluster reflects the degree of drift of the existing state, the emergence of a new cluster represents the introduction of a new pattern, and the extinction of a cluster signifies the disappearance of an old pattern. For example, suppose that from time T-1 to time T, bipartite graph matching reveals three persistent clusters with matching weights, i.e., matching costs, of 0.2, 0.3, and 0.4, respectively. Then, the total cost representing state drift is... This is the sum of these three values, which is 0.9. Simultaneously, at time T, a newly formed cluster that did not exist at time T-1 was discovered. The average distance from all points within this cluster to its centroid was calculated, yielding an intra-cluster dispersion of 1.5. Therefore... A value of 1.5 reflects the compactness of the newly emerging state; greater dispersion indicates a more unstable transition state. Furthermore, it was found that one cluster disappeared at time T-1. We need to find the intra-cluster dispersion of this disappeared cluster at time T-1, assuming its value is 1.1. This is equal to 1.1, representing the stability of the vanished state before it vanished. Using preset weighting coefficients, such as α = 0.4, β = 0.4, and γ = 0.2, the energy increment is calculated. It is 1.18.

[0065] In an optional embodiment, determining that the dehydrator's operating state has changed when the cumulative sum of the predicted innovation sequence and the control chart statistic of the Kalman filter model exceed a preset control limit for N consecutive sampling points includes: obtaining the predicted innovation sequence and its standard deviation using the Kalman filter model. The cumulative control chart statistic is calculated using the following formula.

[0066] ;

[0067] in, Let be the cumulative sum and control chart statistic at time t, and let k be the reference value. , for Accumulated data over time and control chart statistics. For the predicted information at time t;

[0068] When cumulative sum control chart statistics When five consecutive sampling points exceed the control limit H, the dehydrator's operating state is determined to have changed. The control limit H is... .

[0069] Setting a threshold directly for the raw energy value can easily lead to false alarms due to occasional noise spikes. However, the cumulative sum control chart, by accumulating deviations exceeding normal fluctuations, can effectively amplify weak change signals until the cumulative effect is sufficiently significant, thus improving detection accuracy. For example, suppose that by applying Kalman filtering to the evolutionary energy sequence during a stable operation period, a predicted innovation sequence with a mean of 0 and a standard deviation σ of 0.2 is obtained. The reference value k is 0.1, and the control limit H is 1.0. During monitoring, suppose that at time t-1, the cumulative sum statistic... The value is 0.8. At time t, the new evolutionary energy value, after being processed by Kalman filtering, yields the predicted innovation. The value is 0.4. Therefore, the new cumulative control chart statistic is... The calculated value is 1.1. Since 1.1 is greater than 0, Updated to version 1.1. At this point, The value 1.1 has exceeded the control limit H of 1.0. However, this is only the first time it has exceeded the limit, and no alarm will be triggered immediately. At the next time t+1, assuming the news feed is 0.35, then... It equals 1.35, which is still greater than 1.0. This process continues if the calculated values ​​are consecutive... , , If the values ​​are, for example, 1.5, 1.7, and 1.9, all greater than 1.0, then by time t+4, the statistics of five consecutive sampling points exceed the control limits. At this point, the judgment condition is met, and an alarm is issued.

[0070] The implementation principle of the data analysis-based dehydrator operating status detection method of this invention is as follows: First, by defining distance and local structural entropy based on the covariance matrix of the k-nearest neighbor set, the density clustering process is improved, which can effectively cope with the characteristic data of uneven density and blurred boundaries generated by the dehydrator under complex operating conditions, and improve the accuracy of clustering. Second, by constructing a bipartite graph between time window cluster sets and solving the minimum weight matching, the relationship of data cluster changes in the time dimension is established, and the maintenance, creation, and extinction processes of clusters can be quantitatively identified. Furthermore, the system state transition index integrates the matching cost of persistent clusters, the dispersion of newly created clusters, and the survival information of extinct clusters, transforming complex cluster changes into an indicator that is highly sensitive to changes in system stability. In this way, this invention can suppress the interference of random noise, identify weak and persistent trends caused by real changes in state, and thus improve the timeliness and reliability of state change detection.

[0071] An embodiment of the data analysis-based dehydrator operating status detection system provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the data analysis-based dehydrator operating status detection method in the above embodiment.

[0072] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A data analysis-based dewatering machine working state detection method, characterized in that, The method comprises the following steps: acquiring multi-dimensional vibration signals of a dewatering machine during operation, dividing the vibration signals into continuous time windows with an overlap rate, extracting features of data in each time window to obtain a feature vector set; The distance is determined based on an inverse matrix of a covariance matrix of a k-neighbor point set of each feature vector, and a neighborhood radius of density clustering is determined according to the distance; a local structure entropy is obtained based on a Shannon entropy of a normalized eigenvalue of the covariance matrix of the k-neighbor point set, and a feature vector with a local structure entropy lower than a first threshold value and a number of data points in the neighborhood higher than a minimum number threshold value is taken as a core point; the feature vectors in the current time window are subjected to density clustering by using the core point and the neighborhood radius to obtain a current cluster set; a bipartite graph is constructed between the current cluster set and a last time window cluster set, and a weight of an edge of the bipartite graph is determined by a weighted sum of a point set intersection ratio of two clusters, a distance between the point sets and a distance between cluster centers; a mapping relationship of the clusters is established by solving a minimum weight matching problem of the bipartite graph, and a persistent cluster, a newborn cluster and a vanished cluster are identified; The system state transition index is calculated, comprising: ； wherein, is the system state transition index of the current time window, is the system state transition index of the previous time window, and λ is a forgetting factor, is the energy increment of the current time window; the energy increment is obtained by a weighted sum of the matching cost of all persistent clusters, the intra-cluster dispersion of all nascent clusters, and the historical survival duration of all extinct clusters. A Kalman filtering model of a time sequence of the system state transition index is established, and when a cumulative sum control chart statistic of a predicted innovation sequence of the Kalman filtering model continuously exceeds a preset control limit for N sampling points, it is determined that the working state of the dewatering machine changes.

2. The data analysis based working condition detection method of a dewatering machine according to claim 1, characterized in that, The distance is determined based on an inverse matrix of a covariance matrix of a k-neighbor point set of each feature vector, and a neighborhood radius of density clustering is determined according to the distance, comprising: For the feature vector , compute the covariance matrix of its k-nearest neighbors set ; Any two eigenvectors The distance between any two eigenvectors is calculated by the following equation, The distance between any two eigenvectors is calculated by the following equation, ； wherein, is a feature vector is a distance between , is is a transpose of a difference vector between , is is a covariance matrix of is is an inverse matrix of is is an inverse matrix of a covariance matrix of k-nearest points of The distance of each point in the data set to its kth neighbor point is calculated, all distance values are arranged in ascending order to draw a k-dist curve, and a distance value corresponding to an inflection point of the curve is selected as the neighborhood radius of density clustering.

3. The data analysis based working condition detection method of a dewatering machine according to claim 1, characterized in that, The local structure entropy is obtained based on a Shannon entropy of a normalized eigenvalue of the covariance matrix of the k-neighbor point set, comprising: computing eigenvectors covariance matrix of the k nearest neighbors d eigenvalues ; normalizing the eigenvalues to obtain P1, P2, P d ; The local structure entropy of the formula is calculated by the following formula ,​ ； wherein, is the normalized value for the jth feature value.

4. The data analysis based working condition detection method of a dewatering machine according to claim 1, characterized in that, The energy increment is obtained by a weighted sum of a matching cost of all persistent clusters, an intra-cluster dispersion of all newborn clusters and a historical survival duration of all vanished clusters, comprising: Summing the matching cost of all persistent clusters in the bipartite minimum weight matching yields ; calculating the average Euclidean distance of all points within each new-born cluster to the centroid, said average Euclidean distance as the intra-cluster dispersion, summing up the intra-cluster dispersions of all new-born clusters to obtain ; obtaining the intra-cluster dispersion of each dying cluster in the last time window, and summing up the intra-cluster dispersions of all dying clusters to obtain the total intra-cluster dispersion of all dying clusters in the last time window ; Energy increment is calculated by the following equation, ； wherein, , , are preset weighting coefficients.

5. The data analysis based working condition detection method of a dewatering machine according to claim 1, characterized in that, When the cumulative sum control chart statistic of the predicted innovation sequence of the Kalman filtering model continuously exceeds the preset control limit for N sampling points, it is determined that the working state of the dewatering machine changes, comprising: The Kalman filter model is used to obtain the predicted innovation sequence and its standard deviation ; The cumulative sum control chart statistic is calculated by the following formula, ； wherein is the cumulative sum control chart statistic at time t, and the reference value k is , is cumulative and control chart statistics at time t, is the predicted innovation at time t; When the accumulated and controlled chart statistics If the continuous 5 sampling points are greater than the control limit H, it is determined that the working state of the dewatering machine changes, wherein the control limit H is .

6. The data analysis based working condition detection method of a dewatering machine according to claim 1, characterized in that, The multi-dimensional vibration signals of the dewatering machine during operation are acquired, comprising: A three-axis acceleration sensor is installed on a bearing seat of the dewatering machine to collect vibration acceleration signals of the dewatering machine in X, Y and Z directions.

7. The data analysis based working condition detection method of a dewatering machine according to claim 6, characterized in that, The features of data in each time window are extracted, comprising: The time domain features and frequency domain features of the vibration acceleration signals in X, Y and Z directions in each time window are calculated respectively, the time domain features include root mean square, kurtosis, margin and peak value, and the frequency domain features include spectral entropy, spectral kurtosis and center of gravity frequency.

8. The data analysis based working condition detection method of a dewatering machine according to claim 1, characterized in that, The overlap rate is 50%.

9. A data analysis based working condition detection system for a dehydrator, characterized in that, The method comprises a memory and a processor, the memory stores computer instructions, and the processor executes the computer instructions to implement the dewatering machine working state detection method based on data analysis according to any one of claims 1-8.

Images