CN-122015965-A - Online identification method and system for cavitation erosion and abrasion combined damage of centrifugal pump

CN122015965ACN 122015965 ACN122015965 ACN 122015965ACN-122015965-A

Abstract

The invention discloses a method and a system for identifying cavitation erosion abrasion combined damage of a centrifugal pump on line, wherein the method comprises the steps of obtaining a multi-mode feature vector set of the centrifugal pump; according to the multi-mode feature vector set, independent contributions of cavitation damage and abrasion damage are subjected to preliminary decomposition by using a preset signal overlap separation model to obtain potential feature distribution of two damage types, if a cross overlap area exists in the potential feature distribution, whether a unique mode of composite damage exists or not is judged, a classified damage class label is obtained, a damage grading evaluation matrix is constructed according to the classified damage class label and by combining indirect features related to surface roughness and pressure pulsation abnormality caused by flow field change, damage severity of different stages is determined, a preset threshold range is dynamically updated according to the damage severity and feature evolution trend, a judgment basis required by real-time identification is obtained, and differentiation of cavitation damage, abrasion damage and composite damage is completed.

Inventors

HAN XIANGDONG
YU FANGYAN
CHEN LIANG

Assignees

南昌理工学院

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (8)

1. An online identification method for cavitation-abrasion combined damage of a centrifugal pump, which is characterized by comprising the following steps: acquiring a multi-mode feature vector set of a centrifugal pump; according to the multi-mode feature vector set, performing preliminary decomposition on independent contributions of cavitation damage and abrasion damage by using a preset signal overlap separation model, and obtaining potential feature distribution of two damage types; If a cross overlapping area exists in the potential feature distribution, further dividing the potential feature distribution by a multidimensional feature clustering method, judging whether a unique mode of composite damage exists, and obtaining a classified damage class label; Aiming at the classified damage class labels, constructing a damage grading evaluation matrix by combining indirect features related to surface roughness and pressure pulsation abnormality caused by flow field change, and determining the damage severity of different stages; And dynamically updating a preset threshold range according to the severity degree of the damage and the characteristic evolution trend, acquiring a judgment basis required by real-time identification, and completing the differentiation of cavitation damage, abrasion damage and composite damage.
2. The method of claim 1, wherein the method of obtaining a set of multi-modal feature vectors for a centrifugal pump comprises: Collecting vibration signals, pressure pulsation data and acoustic emission wave information when the centrifugal pump operates through a plurality of sensors, and performing synchronous time stamp marking on the collected data to obtain a primarily fused multi-mode data set; And extracting the characteristics of the multi-mode data set by adopting a frequency domain analysis technology, separating out broadband energy distribution in vibration signals, periodic fluctuation in pressure pulsation and high-frequency burst pulse in acoustic emission waves, and determining a characteristic vector set of each mode signal.
3. The method of claim 1, wherein the method for obtaining potential feature distributions of two lesion types by initially decomposing independent contributions of cavitation and erosion lesions using a predetermined signal overlap-separation model based on a set of multi-modal feature vectors comprises: ; In the formula, At a frequency of Then, the damage quantization index of the jth order mode of the ith measuring point, In order to construct the imaginary part of the j-th order complex mode shape of the ith degree of freedom under the damaged state, For the imaginary part of the j-th order complex mode shape of the ith degree of freedom, As a total number of degrees of freedom, Is the natural frequency of the j-th order mode after damage, The natural frequency of the j-th order mode in the healthy state is shown, d is the damage state, In the state of being not damaged, In order to realize the j-th order mode of the structure under the damaged state, the complex amplitude of the i-th free point comprises amplitude and phase information, The complex amplitude of the ith free point is the j-th order mode of the structure in an undamaged state, and comprises amplitude and phase information.
4. The method of claim 1, wherein if there is a cross overlap region in the potential feature distribution, further dividing the potential feature distribution by a multidimensional feature clustering method, determining whether there is a unique pattern of the composite lesion, and obtaining the classified lesion class label comprises: Firstly, acquiring distribution information of potential features from a data set, and primarily identifying a distribution intersection and an overlapping area to obtain a cross overlapping distribution range; Secondly, for the cross overlapped distribution range, adopting a multi-dimensional feature extraction mode, and carrying out subdivision processing on the data by a clustering method to obtain a preliminary subdivision data set; analyzing the related modes of the composite damage according to the preliminary subdivision data set, judging whether the expression form of the unique mode exists or not, and determining the characteristic set of the composite damage; If a significant unique mode exists in the feature set of the composite damage, classifying the feature set to obtain preliminary classification of classification results; And fifthly, according to the preliminary division of the classification result and the definition standard of the damage category, generating a corresponding category label, and determining the attribution of the final damage category.
5. The method of claim 1, wherein the method for constructing a lesion classification evaluation matrix and determining the severity of lesions at different stages for classified lesion class labels by combining indirect features related to surface roughness with flow field variation-induced pressure pulsation anomalies comprises: constructing a feature vector group according to the damage classification label, the surface roughness feature and the pressure pulsation abnormality; The surface roughness characteristics and the pressure pulsation abnormal amplitude values are normalized to obtain a standardized characteristic vector group; adopting a main component analysis dimension reduction processing standardized feature vector group to obtain a dimension reduction feature vector; grouping and clustering the dimension reduction feature vectors according to the damage classification labels to obtain feature cluster centers corresponding to a plurality of damage categories; determining the nearest cluster center by calculating Euclidean distance from the dimension reduction feature vector to each cluster center to obtain the attribution of the preliminary damage category; Establishing two-dimensional evaluation coordinates according to the surface roughness characteristic value and the pressure pulsation abnormal amplitude value, and mapping the dimension reduction characteristic vector to the two-dimensional evaluation coordinates; Pre-establishing a damage grading evaluation matrix, wherein a matrix row index corresponds to a roughness characteristic segment, a matrix column index corresponds to a pressure pulsation abnormal segment, a matrix element stores a severity level, and a random forest model is adopted for training to obtain the grading evaluation matrix; and obtaining the damage severity level by searching the matrix row index and matrix array index positions corresponding to the dimension reduction feature vector in the two-dimensional evaluation coordinates.
6. The method of claim 5, wherein establishing two-dimensional evaluation coordinates from the surface roughness feature values and the pressure pulsation anomaly amplitude values, and mapping the dimension-reduced feature vector to the two-dimensional evaluation coordinates comprises: ; In the formula, In order to be a roughness feature matrix, In order to be a pressure pulsation characteristic matrix, For the horizontal splicing operation, For principal component analysis, before reservation The main components of the composition are as follows, For the PCA projection matrix, For the two-dimensional evaluation of the coordinate matrix of the output, Is the first The coordinates of the individual samples are then used, As a dimension of the roughness characteristics, Is the pressure pulsation characteristic dimension.
7. The method of claim 1, wherein dynamically updating the predetermined threshold range according to the severity of the damage and the characteristic evolution trend to obtain a decision basis required for real-time identification, and the method for distinguishing the cavitation damage, the abrasion damage and the composite damage comprises the following steps: Firstly, acquiring damage related data through a sensor, primarily recording the damage degree and evolution trend, and obtaining a continuous data stream of damage change by adopting a time sequence analysis method; Step two, according to the continuous data flow of the damage change, combining with a preset threshold value, implementing a dynamic adjustment strategy, and if the data flow exceeds the preset threshold value range, triggering a threshold value updating mechanism to determine a new threshold value range; thirdly, aiming at the updated threshold range, acquiring data input required by real-time identification, classifying damage features by using a support vector machine algorithm, and judging whether the damage belongs to cavitation damage or abrasion damage; Step four, according to the classification result, combining with the judgment basis, if the classification result shows that multiple characteristics overlap, further analyzing the damage type, and determining whether the damage is a composite damage or not; acquiring a basis of accurate distinction through an analysis result of the damage types, and if the characteristics of the single damage type are obvious, classifying the single damage type as the corresponding damage type directly to acquire a final recognition result; and step six, generating a structural data record of the damage type according to the final identification result, and storing the structural data record into a database by adopting a data storage technology to acquire basic data of subsequent analysis.
8. An online recognition system of cavitation erosion and abrasion combined damage of a centrifugal pump, which is used for realizing the method of any one of claims 1-7, and is characterized by comprising an extraction module, a damage contribution decomposition module, a composite damage recognition module, a damage grading evaluation module and a dynamic threshold updating module; The extraction module is used for obtaining a multi-mode feature vector set of the centrifugal pump; the damage contribution decomposition module is used for carrying out preliminary decomposition on independent contributions of cavitation damage and abrasion damage by utilizing a preset signal overlap separation model according to the multi-mode feature vector set, so as to obtain potential feature distribution of two damage types; The composite damage identification module is used for further dividing the potential feature distribution by a multidimensional feature clustering method if a cross overlapping area exists in the potential feature distribution, judging whether a unique mode of composite damage exists, and obtaining a classified damage type label; The damage grading evaluation module is used for constructing a damage grading evaluation matrix aiming at classified damage class labels and combining indirect characteristics related to surface roughness and pressure pulsation abnormality caused by flow field change to determine the damage severity of different stages; the dynamic threshold updating module is used for dynamically updating a preset threshold range according to the severity degree of damage and the characteristic evolution trend, acquiring a judgment basis required by real-time identification and completing the differentiation of cavitation damage, abrasion damage and composite damage.

Description

Online identification method and system for cavitation erosion and abrasion combined damage of centrifugal pump Technical Field The invention belongs to the technical field of damage identification, and particularly relates to a cavitation-abrasion combined damage on-line identification method and system for a centrifugal pump. Background As core equipment in petrochemical industry, electric power, water treatment and other industries, once cavitation and abrasion combined damage occurs, the centrifugal pump often causes quick failure of the impeller and even causes complete machine parking or safety accidents, so that the centrifugal pump has extremely high value for accurately identifying the damage state in real time. The existing identification method is designed separately for cavitation erosion or abrasion, so that the actual situation that two types of damages occur simultaneously and are mutually promoted is difficult to deal with, particularly when a medium containing solid particles flows in a pump, the impact of the particles on the surface of an impeller is aggravated by the bubble breakage generated by cavitation erosion, and the cavitation erosion is easier to go deep due to the fact that the surface protection layer is damaged by the particle erosion, so that the damage process is obviously accelerated and is difficult to explain by a single mechanism. In actual operation, cavitation and erosion feature signals overlap highly, and similar broadband energy increases occur in vibration, pressure pulsation and acoustic emission, but correspond to completely different lesion morphologies and severity. More complicated, new composite characteristics can be generated after two kinds of damage are overlapped, such as honeycomb pits and directional flushing grooves on the surface of the impeller, so that frequent false alarm or missing alarm is judged based on the threshold established by single damage. When the damage progresses to the middle and late stages, the surface roughness is increased further to change the flow field, so that stronger pressure pulsation and bubble collapse are induced, a vicious circle is formed, and the damage degree is increased from slight rapid to severe in a short time. Therefore, how to accurately distinguish the composite damage generated by simple cavitation erosion, simple abrasion and superposition of the cavitation erosion and the abrasion under the condition that cavitation erosion and abrasion signals are mutually interfered and damage characteristics continuously evolve along with time, and to realize reliable grading screening on the damage of different stages such as microscopic crack expansion, material peeling, surface roughness change and the like becomes a key problem for restricting the long-term safe operation of the centrifugal pump. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a cavitation-abrasion combined damage online identification method and system for a centrifugal pump, which remarkably improve the accuracy and timeliness of the health diagnosis of the centrifugal pump. In order to achieve the above object, the present invention provides the following solutions: an on-line identification method for cavitation-abrasion combined damage of a centrifugal pump, comprising the following steps: acquiring a multi-mode feature vector set of a centrifugal pump; according to the multi-mode feature vector set, performing preliminary decomposition on independent contributions of cavitation damage and abrasion damage by using a preset signal overlap separation model, and obtaining potential feature distribution of two damage types; If a cross overlapping area exists in the potential feature distribution, further dividing the potential feature distribution by a multidimensional feature clustering method, judging whether a unique mode of composite damage exists, and obtaining a classified damage class label; Aiming at the classified damage class labels, constructing a damage grading evaluation matrix by combining indirect features related to surface roughness and pressure pulsation abnormality caused by flow field change, and determining the damage severity of different stages; And dynamically updating a preset threshold range according to the severity degree of the damage and the characteristic evolution trend, acquiring a judgment basis required by real-time identification, and completing the differentiation of cavitation damage, abrasion damage and composite damage. Preferably, the method for acquiring the multi-modal feature vector set of the centrifugal pump comprises the following steps: Collecting vibration signals, pressure pulsation data and acoustic emission wave information when the centrifugal pump operates through a plurality of sensors, and performing synchronous time stamp marking on the collected data to obtain a primarily fused multi-mode data set; And extracting the characteristics of the mu