A state early warning method and system for a centrifuge

Abstract

The present application relates to the technical field of electric data processing, and more particularly, to a state early warning method and system for a centrifugal machine, comprising: collecting vibration signals and rotating speeds of the centrifugal machine, and obtaining vibration amplitudes of one rotation frequency components corresponding to the rotating speeds in the vibration signals; for each rotating speed, collecting vibration amplitudes of multiple one rotation frequency components under the rotating speed as a sample set; for any rotating speed, calculating a fluctuation entropy corresponding to the rotating speed, the fluctuation entropy being used to represent an expected dispersion degree of the vibration amplitudes under a healthy state. The present application dynamically adjusts parameters of a DBSCAN clustering algorithm by combining the fluctuation entropy related to the rotating speed and a deviation factor reflecting an actual dispersion degree of data, which makes the algorithm effectively eliminate abnormal data under different working conditions, thereby establishing a more accurate and robust health state baseline, and significantly improving the accuracy of fault early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric data processing. More particularly, the present application relates to a state early warning method and system for a centrifuge. BACKGROUND

[0002] As a key equipment widely used in the fields of bioengineering, medical diagnosis and material science, the stability and safety of centrifuge in operation are of great importance. Rotor imbalance is one of the most common and destructive failure modes of centrifuge, which can cause severe vibration during equipment operation, not only affecting the separation effect, but also possibly leading to rotor damage, sample loss and even safety accidents.

[0003] Through comprehensive analysis and processing of the vibration signals of the centrifuge in the actual operation process, not only the current dynamic working characteristics of the equipment can be accurately reflected, but also reliable basis can be provided for realizing real-time monitoring of the health state and early warning of potential faults. Among them, especially the one-fold rotational frequency vibration amplitude, which has a high correlation between the rotor imbalance state, can intuitively represent the deviation of the rotor force and motion, and is therefore often considered as the core monitoring index for judging the running state, stability and potential risk of the centrifuge.

[0004] In order to accurately warn, a reliable normal operation vibration baseline needs to be established based on a large amount of historical vibration data collected under the health state of the equipment. However, in the actual collection process, abnormal values or noise points caused by instantaneous disturbance, measurement error or concurrent fault and other factors will inevitably be mixed in the historical data. These abnormal data will seriously interfere with the establishment of the baseline model, and if not removed, will lead to the inaccuracy of the calculated normal state range, thereby causing false positives or false negatives of the equipment state warning system, reducing the accuracy and reliability of the warning. SUMMARY

[0005] The present application provides a state early warning method and system for a centrifuge, which aims to solve the problem that in the related art, abnormal data in the historical data will seriously interfere with the establishment of the baseline model, and if not removed, will lead to the inaccuracy of the calculated normal state range, thereby causing false positives or false negatives of the equipment state warning system.

[0006] In a first aspect, the present application provides a method for early warning of a centrifuge, comprising: collecting vibration signals and rotation speeds of the centrifuge, and obtaining vibration amplitudes of one rotation frequency components corresponding to the rotation speeds in the vibration signals; for each rotation speed, collecting vibration amplitudes of a plurality of one rotation frequency components at the rotation speed as a sample set; for any rotation speed, calculating a fluctuation entropy corresponding to the rotation speed, the fluctuation entropy being used to represent an expected dispersion degree of the vibration amplitudes in a healthy state, calculating a ratio of an actual dispersion degree of the sample set to the fluctuation entropy to obtain a deviation factor of the sample set at the rotation speed; determining a neighborhood radius for DBSCAN clustering, the neighborhood radius being positively related to a product of the fluctuation entropy and a first adjustment coefficient thereof, and being negatively related to a product of the deviation factor and a second adjustment coefficient thereof; performing DBSCAN clustering on the sample set, and removing noise points in a clustering result from the sample set to obtain a pure data subset; wherein the first adjustment coefficient and the second adjustment coefficient are obtained based on a particle swarm algorithm; calculating a statistical value of the pure data subset as a baseline vibration amplitude at the corresponding rotation speed, and judging a device state of the centrifuge according to a real-time rotation speed and the corresponding baseline vibration amplitude. By introducing the fluctuation entropy related to the rotation speed and the deviation factor reflecting the actual dispersion of the data, the neighborhood radius of the DBSCAN algorithm is dynamically adjusted. This adaptive parameter adjustment enables the algorithm to accurately identify and remove abnormal data points according to the data distribution characteristics at different rotation speeds, thereby establishing a more pure and reliable vibration baseline, and significantly improving the accuracy and robustness of the centrifuge state early warning.

[0007] Further, the calculation of the fluctuation entropy corresponding to the rotation speed comprises: taking a product of a rotation speed energy factor of the rotation speed and a basic fluctuation entropy as the fluctuation entropy corresponding to the rotation speed.

[0008] Further, the calculation of the actual dispersion degree of the sample set comprises: calculating a quartile range of the sample set as the actual dispersion degree. The quartile range is used as the actual dispersion degree, and since the quartile range has strong robustness to abnormal values in the data, it can more truly reflect the dispersion degree of the main part of the sample data, avoiding overestimation of the volatility of the data due to the existence of a few abnormal points, thereby improving the calculation accuracy of the deviation factor.

[0009] Further, the first adjustment coefficient and the second adjustment coefficient are obtained based on a particle swarm algorithm, including: inputting a random combination of the first adjustment coefficient and the second adjustment coefficient as a particle into the particle swarm algorithm for optimization, stopping the search process when a fitness function value converges or reaches a maximum iteration number, and obtaining the optimal first adjustment coefficient and the second adjustment coefficient, that is, the optimal adjustment coefficient of the DBSCAN clustering.

[0010] Further, the fitness function value is constructed according to the following formula: ; in the formula, is the first adjustment coefficient, is the second adjustment coefficient, represents the number of normal points divided into the core cluster after clustering using the particle parameter, represents the total number of normal points in the labeled data, represents the number of abnormal points divided into the core cluster after clustering using the particle parameter, represents the total number of abnormal points in the labeled data, represents the total number of points in the core cluster. The function ingeniously combines multiple dimensions such as recall rate of normal points, exclusion of abnormal points and purity of core cluster, and constructs a comprehensive evaluation index. Compared with single pursuit of classification accuracy, the fitness function value of the application can more comprehensively evaluate the quality of the clustering result, guide the optimization algorithm to find a parameter combination that can achieve the best balance between correctly identifying normal data and effectively excluding abnormal data, so that the final model is more practical and reliable.

[0011] Further, the vibration amplitudes of the plurality of one-fold rotational frequency components at the rotational speed are collected as a sample set, including: obtaining vibration signals collected continuously for a period of time at the rotational speed, dividing the vibration signals for a period of time into equal-length time windows, and performing fast Fourier transform on the signals in each time window to extract the vibration amplitudes of the one-fold rotational frequency components, thereby obtaining a sample set composed of the vibration amplitudes of the plurality of one-fold rotational frequency components. This method can quickly obtain a set of vibration amplitude samples with statistical significance at a single stable rotational speed. Compared with scattered and random single measurement, this method ensures the consistency and representativeness of the sample data, and provides a high-quality data basis for subsequent accurate dispersion calculation and density clustering analysis.

[0012] Further, judging the equipment state of the centrifuge comprises: determining a safety interval of the vibration amplitude corresponding to the rotation speed, the safety interval being a mean value of all vibration amplitudes in the pure data subset 3 times of a standard deviation of all vibration amplitudes in the data subset; obtaining the rotation speed and the vibration amplitude of the centrifuge at the current moment, and detecting the equipment state of the centrifuge based on whether the vibration amplitude of the centrifuge at the current moment is in the safety interval.

[0013] Further, detecting the equipment state of the centrifuge comprises: if the vibration amplitude of the centrifuge at the current moment is not in the safety interval of the vibration amplitude corresponding to the rotation speed, alarming and recording.

[0014] Further, calculating the statistical value of the pure data subset comprises: calculating a mean value or a median of the pure data subset.

[0015] The second aspect of the present application also provides a state early warning system for a centrifuge, comprising a processor and a memory, the memory storing a computer program, and the processor executing the computer program to realize the state early warning method for the centrifuge according to any one of the above.

[0016] Beneficial effects: by combining the rotation speed related fluctuation entropy and the deviation factor reflecting the actual dispersion degree of data, the parameters of the DBSCAN clustering algorithm are dynamically adjusted, which makes the algorithm effectively eliminate abnormal data under different working conditions, thereby establishing a more accurate and robust health state baseline, and significantly improving the accuracy of fault early warning. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flow chart illustrating the centrifuge state monitoring according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0019] As Figure 1 shown, S101: collecting the running signals of the centrifuge.

[0020] First, the operating parameters of the centrifuge during the working process are collected. Specifically, by deploying an acceleration sensor on the bearing seat of the centrifuge, a vertical or horizontal vibration signal sequence is collected in real time. At the same time, by connecting to the speed sensor of the centrifuge main shaft or from the drive controller, the real-time speed of the centrifuge is synchronously acquired. In order to extract the characteristics related to unbalanced vibration from the collected vibration signals, fast Fourier transform (FFT) is applied to the vibration signals in each time window collected, to obtain the frequency domain representation of the vibration signals. Fast Fourier transform is an efficient algorithm for converting time domain signals to frequency domain signals, thereby revealing the frequency components of the signal. From the frequency spectrum, the fundamental frequency (i.e. one times the rotational frequency, marked as 1X) component corresponding to the real-time speed is identified and extracted, and its amplitude is obtained. The amplitude of 1X is a key indicator for measuring the unbalance state of the rotor.

[0021] S102: Construct the baseline vibration amplitude of each speed.

[0022] In order to eliminate the influence of the difference in the highest speed of different models of centrifuges, the speed of the centrifuge is first normalized to eliminate the dimensional difference and make the model more universal.

[0023] First, before the state warning of the centrifuge, the operating state data of the centrifuge in a period of time before the current time is obtained, and the operating state of the centrifuge in a period of time is obtained in turn at a plurality of preset stable speed points . For each stable speed , , the i-th speed, the vibration signal is continuously collected at this speed for a period of time. The signal is divided into equal length time windows, and the amplitude of the one times rotational frequency component is extracted by performing fast Fourier transform (FFT) on the signal in each time window. Thus, at the i-th speed , a sample set containing vibration amplitude measurement values is obtained. In this embodiment, takes the value of 10, and in other embodiments, the value of which can be 5 or 7, etc., which can be adjusted according to the implementation.

[0024] Under different rotational speed conditions, the vibration characteristics of the centrifuge are different, which leads to the difference in the data distribution and density of the vibration amplitude sample set collected from each stable rotational speed point. Specifically, the normal vibration amplitude under high rotational speed may be larger than the abnormal value under low rotational speed. As a density-based clustering method, the core parameters of the DBSCAN algorithm, i.e., the neighborhood radius and the minimum number of points, are crucial to the result. If a global and unified Eps value is used when processing data under all rotational speeds, the radius may be too large under the rotational speed with compact data distribution, resulting in the failure to effectively identify abnormal points; while under the rotational speed with sparse data distribution, the radius may be too small, causing a large number of normal data points to be misjudged as noise. In order to solve this problem and improve the adaptability of the model to different conditions and the robustness of abnormal value identification, the DBSCAN algorithm must be optimized. Specifically, for each sample set under a rotational speed, an adaptive method should be used to determine its exclusive clustering radius, and the specific process is as follows.

[0025] Firstly, for any rotational speed, the fluctuation entropy under the rotational speed is constructed to represent the expected dispersion degree of the vibration amplitude data of the healthy centrifuge under the rotational speed. The construction process of the fluctuation entropy is as follows: based on the principle of rotating machinery dynamics, the higher the rotational speed, the greater the kinetic energy contained in the system, and the more intense the inherent random vibration fluctuation. To quantify this physical property, first define a rotational speed energy factor , . for the i th rotational speed, for the maximum rotational speed, the rotational speed energy factor is proportional to the square of the rotational speed , and a slight increase in rotational speed will lead to a significant increase in the energy level of the system, wherein on the centrifuge main machine, there is usually a manufacturer's nameplate or label that clearly indicates the highest rotational speed, which is the maximum rotational speed.

[0026] Based on the rotational speed energy factor, the fluctuation entropy is further constructed, and its construction formula is: . is the fluctuation entropy of the i th rotational speed, representing the dispersion degree of the normal vibration amplitude sample set under the healthy state at the current rotational speed, is the basic fluctuation entropy, representing the minimum vibration fluctuation of the centrifuge at the lowest rotational speed, and its value range can be set to 0.01-0.05, is the rotational speed energy factor of the i th rotational speed.

[0027] is introduced to quantify the difference between the actual dispersion of the currently collected sample set and its dispersion under the healthy state. To ensure the robustness of the calculation and avoid the interference of abnormal values, the interquartile range is used to calculate the actual dispersion of the sample set , ; wherein, is the actual dispersion of the sample set ; is the third quartile of the sample set ; is the first quartile of the sample set ; is the sample set of vibration amplitude collected at the rotational speed . The deviation factor is the normalized value of the ratio of the actual dispersion and the fluctuation entropy. When the value of the deviation factor is close to 1, it indicates that the dispersion degree of the measured data set is basically consistent with the dispersion under the healthy state. When the deviation factor is much greater than 1, it means that a large number of abnormal values caused by concurrent faults may be mixed in the sample set.

[0028] Then, based on the deviation factor corresponding to the sample set at the rotational speed, a neighborhood radius is generated and clustering is performed. Specifically, a dynamic adaptive DBSCAN neighborhood radius is generated in combination with the fluctuation entropy and the deviation factor. . In the formula, is the neighborhood radius finally used for the DBSCAN algorithm at the rotational speed ; is the first adjustment coefficient, is the second adjustment coefficient, is the fluctuation entropy of the i-th rotational speed, is the deviation factor of the sample set . The construction logic is that the core of the dynamic neighborhood radius is , which ensures that the value of can be adaptively increased as the rotational speed increases. The correction term plays a key adjustment role: when the measured data fluctuation is normal, a larger neighborhood radius should be used; when the measured data fluctuation is abnormally large, the value of is smaller, which will make the value of shrink to its basic value . This shrinkage effect makes DBSCAN use a more strict and smaller neighborhood radius when facing a severely contaminated data set, so as to more effectively identify those abnormal points that cause the overall dispersion to increase as noise.

[0029] Finally, for each stable rotational speed , the dynamic neighborhood radius calculated at the rotational speed is used to perform DBSCAN clustering on the sample set . The algorithm will divide the data points into core clusters and noise points. All data marked as noise points are removed to obtain a pure data subsetFinally, the mean or median of the amplitude of vibration at the rotational speed is calculated as the baseline amplitude of vibration .

[0030] It should be noted that the values of the adjustment coefficients and directly affect the effect of DBSCAN clustering. In order to determine the optimal combination thereof, an optimization method based on historical labeled data can be used. Specifically, historical data containing normal and abnormal labels at the rotational speed within a period of time between the current time is obtained, and a random combination of the values of and is used as a particle in a particle swarm optimization algorithm. The dynamic neighborhood radius of each particle is calculated using the above method, DBSCAN clustering is performed, and the accuracy of the clustering result (i.e., clustering normal points into core clusters and identifying abnormal points as noise) is maximized as the fitness function value . Then, the particle swarm optimization algorithm is used to iteratively search for and , and the search is stopped when the fitness function value converges or the maximum number of iterations is reached. The and of the optimal particle obtained are the optimal parameter values, wherein the particle corresponding to the maximum fitness function value is the optimal particle.

[0031] As mentioned above, the fitness function value is constructed as follows: . In the formula, represents the number of normal points clustered into core clusters using the particle parameters, represents the total number of normal points in the labeled data, represents the number of abnormal points clustered into core clusters using the particle parameters, represents the total number of abnormal points in the labeled data, represents the total number of points in the core clusters. represents the net benefit of clustering normal points into cores and avoiding clustering abnormal points into cores, and also represents the proportion of normal points correctly clustered as core points. The larger the value, the better, represents the proportion of normal points among all points identified as core points, which can also be referred to as the purity of the clustering cluster. The larger the value, the better.

[0032] S103: For each rotational speed, the alarm threshold at the rotational speed is calculated, and the state of the centrifuge is monitored.

[0033] Specifically, according to the above method, a pure data subset corresponding to each rotation speed can be obtained, and then according to the 3-Sigma principle, a safety interval of the vibration amplitude corresponding to the rotation speed is calculated, the safety interval being the mean value of all vibration amplitudes in the pure data subset 3 times the standard deviation of all vibration amplitudes in the data subset. Finally, during the operation of the centrifuge, the current rotation speed and its baseline vibration amplitude are obtained in real time, the safety interval of the vibration amplitude corresponding to the current rotation speed is calculated, if the baseline vibration amplitude is in the safety interval of the vibration amplitude corresponding to the current rotation speed, it is determined that the equipment is in a normal operating state and no operation is performed; if the baseline vibration amplitude is not in the safety interval of the vibration amplitude corresponding to the current rotation speed, it is determined as an abnormal event and is recorded.

[0034] The present application also provides a state early warning system for a centrifuge. The system comprises a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, realizing a state early warning method for a centrifuge according to the first aspect of the present application.

[0035] The system also comprises a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and thus will not be described here.

[0036] In the present application, the aforementioned memory can be any tangible medium containing or storing a program, which can be used or combined with an instruction execution system, device or apparatus. For example, the computer readable storage medium can be any appropriate magnetic storage medium or magneto-optical storage medium, such as resistive random access memory RRAM (Resistive Random Access Memory), dynamic random access memory DRAM (Dynamic Random Access Memory), static random access memory SRAM (Static Random-Access Memory), enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), high bandwidth memory HBM (High-Bandwidth Memory), hybrid memory cube HMC (Hybrid Memory Cube) and the like, or any other medium that can be used to store desired information and can be accessed by an application, module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device. Any application or module described in the present application can be implemented using computer readable / executable instructions stored or otherwise held by such computer readable medium.

[0037] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation to the scope of the application. It should be noted that for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A status early warning method for a centrifuge, characterized in that, include: Collect the vibration signal and rotation speed of the centrifuge, and obtain the vibration amplitude of the frequency component of the vibration signal that corresponds to the rotation speed; For each rotational speed, the vibration amplitude values ​​of multiple one-time frequency components at that rotational speed are collected as a sample set; For any given rotational speed, calculate the corresponding fluctuation entropy, including: The product of the speed energy factor and the basic fluctuation entropy at that speed is taken as the fluctuation entropy corresponding to that speed; the basic fluctuation entropy represents the minimum vibration fluctuation of the centrifuge at the lowest speed. Rotational speed energy factor satisfy: ; For the i-th rotational speed, This is the maximum speed; Wave entropy is used to characterize the expected dispersion of vibration amplitude under healthy conditions. The ratio of the actual dispersion of the sample set to the wave entropy is calculated to obtain the deviation factor of the sample set at that rotational speed. Determine the neighborhood radius used for DBSCAN clustering; Neighborhood radius satisfy: ; The first adjustment coefficient, This is the second adjustment coefficient. Let i be the fluctuation entropy at the i-th rotational speed. For sample set The deviation factor; DBSCAN clustering is performed on the sample set to remove noisy points from the clustering results, resulting in a clean data subset. The first and second adjustment coefficients are obtained based on the particle swarm optimization algorithm, including: The random combination of the first and second adjustment coefficients is used as particles and input into the particle swarm optimization algorithm for optimization. When the fitness function value converges or reaches the maximum number of iterations, the search process stops, and the optimal first and second adjustment coefficients are obtained, which are the optimal adjustment coefficients of DBSCAN clustering. The fitness function value reflects the accuracy of the clustering results using DBSCAN clustering. The first adjustment coefficient characterizes the degree of influence of fluctuation entropy on the clustering results, and the second adjustment coefficient characterizes the degree of influence of the bias factor on the clustering results. fitness function value satisfy: ； The first adjustment coefficient, This is the second adjustment coefficient. , These represent the number of normal points and the number of outliers that were assigned to the core cluster after clustering using this particle parameter, respectively. , These represent the total number of normal points and the total number of outliers in the labeled data, respectively. This represents the total number of nodes in the core cluster; Calculate the statistical values ​​of a subset of clean data as the baseline vibration amplitude at the corresponding rotational speed, and determine the equipment status of the centrifuge based on the real-time rotational speed and the corresponding baseline vibration amplitude.

2. The status early warning method for a centrifuge according to claim 1, characterized in that, Calculating the actual dispersion of the sample set includes: The interquartile range of the sample set is calculated as the actual dispersion.

3. The status early warning method for a centrifuge according to claim 1, characterized in that, The vibration amplitudes of multiple one-times-revolution frequency components at this rotational speed are collected as a sample set, including: Acquire vibration signals continuously over a period of time at this rotational speed, and then divide the vibration signals over this period into... The signal is divided into several time windows of equal length, and a fast Fourier transform is performed on the signal within each time window to extract the vibration amplitude of the one-time frequency component, resulting in a sample set composed of the vibration amplitudes of multiple one-time frequency components.

4. The status early warning method for a centrifuge according to claim 1, characterized in that, Determining the equipment status of the centrifuge includes: Determine the safe range of vibration amplitude at this rotational speed, where the safe range is the mean of all vibration amplitudes in the pure data subset. Three times the standard deviation of all vibration amplitudes in the data subset; The centrifuge's rotational speed and vibration amplitude are obtained at the current moment, and the centrifuge's equipment status is detected based on whether the vibration amplitude is within the safe range at the current moment.

5. The status early warning method for a centrifuge according to claim 4, characterized in that, The equipment status of the centrifuge is monitored, including: If the current vibration amplitude of the centrifuge is not within the safe range for the vibration amplitude corresponding to the rotation speed, an alarm will be triggered and the data will be recorded.

6. The status early warning method for a centrifuge according to claim 1, characterized in that, Calculating the statistics of the pure data subset includes: Calculate the mean or median of the pure data subset.

7. A status early warning system for a centrifuge, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the status warning method for a centrifuge as described in any one of claims 1-6.

Images