Concrete piston fault prediction method for extracting features by combining clustering idea
Technical Field
The invention relates to the technical field of industrial equipment fault early warning, in particular to a concrete piston fault prediction method for extracting characteristics by combining a clustering idea.
Background
For maintenance of production equipment, the traditional methods mainly comprise two types, one is to maintain after failures occur, but the production is stopped unplanned, and the economic loss is large; the second is maintenance on a fixed schedule, but with high maintenance costs and long down times. And predictive maintenance, namely, performing predictive early warning on the residual life or faults of the key parts of the equipment by analyzing fault historical data and real-time monitoring data, and performing maintenance according to the predictive early warning, so that the unplanned downtime of the equipment is reduced, and the maintenance cost is reduced.
However, there is currently no mature predictive maintenance method. The traditional knowledge-based method is mainly based on the heuristic experience knowledge of relevant experts and operators, qualitatively or quantitatively describes the connection relation, the fault propagation mode and the like among all units in the process, and simulates the reasoning ability of process experts on monitoring through reasoning, deduction and the like after abnormal symptoms appear in equipment, so that the equipment fault early warning and equipment monitoring are automatically completed. The method is based on experience, an accurate mathematical model cannot be established, and the prediction precision and reliability are generally poor, so that the possible future equipment faults are difficult to accurately early warn according to the existing equipment conditions.
According to the equipment fault early warning and state monitoring, the abnormal condition of the equipment is forecasted in time before the equipment really breaks down according to the equipment running rule or the possibility precursor obtained by observation, and corresponding measures are taken, so that the loss caused by the equipment fault can be reduced to the maximum extent. With the increasing scale and complexity of equipment and engineering control systems, it is very urgent and important to monitor and diagnose process abnormalities effectively and timely by reliable state monitoring technology in order to ensure the safety and stability of the production process.
The concrete piston is a key part of the concrete pump truck and is also a consumable part, and the failure of the piston can cause the pump truck to work normally and can cause the normal construction of other supporting equipment in the whole construction site, thereby bringing about great economic loss. The service life of the piston is closely related to the specific working conditions of equipment and the like, real-time working condition data and the like of the pump truck are uploaded to an industrial internet cloud platform through the internet of things, a proper model is built based on accumulated data, and effective prediction and early warning are expected to be made on possible faults of the concrete piston in a certain future working task period, so that operators are reminded to carry out necessary maintenance before construction, and economic loss caused by planned outages is avoided.
At present, the prior art has the following disadvantages:
1. the traditional non-machine learning algorithm cannot predict the concrete piston fault, and once the fault occurs, huge loss is generated, and even the personal safety of workers is damaged.
2. The feature processing method in the prior machine learning algorithm is simple and single, can effectively provide and mine potential information in a non-invasive manner, provides strong features and reduces the prediction accuracy.
Based on the situation, the invention provides a concrete piston fault prediction method for extracting characteristics by combining a clustering idea.
Disclosure of Invention
In order to make up for the defects of the prior art, the invention provides a simple and efficient concrete piston fault prediction method for extracting features by combining a clustering idea.
The invention is realized by the following technical scheme:
a concrete piston fault prediction method for extracting features by combining a clustering idea is characterized by comprising the following steps:
the first step is as follows: the method comprises the following steps that an industrial Internet of things edge calculation box installed on a concrete pump truck is used for collecting working condition data, and the collected working condition data are uploaded to a cloud working condition database;
the second step is that: after the completion condition data is collected, firstly, carrying out visual Pearson correlation analysis among all the features, selecting the features with strong correlation, and carrying out three-clustering on the selected features with strong correlation;
in the second step, after the obtained working condition data are analyzed, the correlation between the engine speed, the oil pump speed and the pumping pressure which are obtained by observation is more than 0.65, so that a new data set is established by three characteristics after the three characteristic working condition data of the engine speed, the oil pump speed and the pumping pressure are copied, and the obtained new data set is subjected to three-clustering by using a K-means clustering algorithm;
the third step: after the clustering result is obtained, respectively calculating contour coefficients for three central points of the trimerization class to obtain three new characteristics for describing the quality of clustering;
under normal conditions, the data change of three characteristics of the engine speed, the oil pump speed and the pumping pressure of the concrete piston data is more consistent, so that the clustering effect is better, and the profile coefficient is larger; if three characteristics of the engine speed, the oil pump speed and the pumping pressure of the concrete piston data are disordered, the clustering effect is poor, and the corresponding profile coefficient is reduced, the concrete piston is predicted to be in failure;
the fourth step: training and predicting data by utilizing LightGBM, XGboost and Catboost algorithms respectively, and finally fusing three prediction results by an averaging method;
in the fourth step, adding new features into a data set consisting of the features with original strong correlation to form a new training set; recording corresponding normal labels and fault labels in a label file of the training set, wherein the normal labels and the fault labels are respectively represented by 0 and 1; wherein 0 indicates that the corresponding piston of the corresponding sample has not failed during the future 2000-square concrete pumping task, and 1 indicates that the corresponding piston of the corresponding sample has failed during the future 2000-square concrete pumping task.
In the first step, the working condition data of the concrete pump truck are acquired by using the sensor and the edge end internet of things computing box, and the acquired working condition data comprise the working time of a piston (the accumulated working time after the piston is newly replaced, a numerical type), the engine speed, the oil pump speed, the pumping pressure, the hydraulic oil temperature, the flow gear, the distribution pressure, the discharge current amount (which is the corresponding working condition value and the numerical type of the pump truck), a low-voltage switch, a high-voltage switch, a stirring overpressure signal, a positive pump, a reverse pump (switching value) and the equipment type (the type and the category of the pump truck).
And in the second step, a KMeans function under a sklern.
When the data set consisting of the engine speed, the oil pump speed and the pumping pressure is subjected to trimerization, the number of clusters, namely the setting of a K value is determined according to multiple tests of the accuracy of prediction, the K value with the best prediction result is selected, and the parameter is set to be KMeans (K is 3).
In the third step, the contour coefficient is calculated for the clustering result by using silouette _ samples under the skearn. metrics package in python.
The invention has the beneficial effects that: the concrete piston fault prediction method for extracting the features by combining the clustering idea is combined with the Pearson correlation coefficient and the K-means clustering method to extract deeper feature information in the features, so that the prediction accuracy and robustness are enhanced, the reliability and accuracy of concrete piston fault early warning are greatly improved, and unnecessary production loss can be reduced.
Drawings
FIG. 1 is a schematic diagram of a concrete piston fault prediction method for feature extraction by combining a clustering idea.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects to be solved by the present invention more apparent, the present invention is described in detail below with reference to the embodiments. It should be noted that the specific embodiments described herein are only for explaining the present invention and are not used to limit the present invention.
The concrete piston fault prediction method for extracting features by combining the clustering idea comprises the following steps:
the first step is as follows: the method comprises the following steps that an industrial Internet of things edge calculation box installed on a concrete pump truck is used for collecting working condition data, and the collected working condition data are uploaded to a cloud working condition database;
the second step is that: after the completion condition data is collected, firstly, carrying out visual Pearson correlation analysis among all the features, selecting the features with strong correlation, and carrying out three-clustering on the selected features with strong correlation;
in the second step, after the obtained working condition data are analyzed, the correlation between the engine speed, the oil pump speed and the pumping pressure which are obtained by observation is more than 0.65, so that a new data set is established by three characteristics after the three characteristic working condition data of the engine speed, the oil pump speed and the pumping pressure are copied, and the obtained new data set is subjected to three-clustering by using a K-means clustering algorithm;
the third step: after the clustering result is obtained, respectively calculating contour coefficients for three central points of the trimerization class to obtain three new characteristics for describing the quality of clustering;
under normal conditions, the data change of three characteristics of the engine speed, the oil pump speed and the pumping pressure of the concrete piston data is more consistent, so that the clustering effect is better, and the profile coefficient is larger; if three characteristics of the engine speed, the oil pump speed and the pumping pressure of the concrete piston data are disordered, the clustering effect is poor, and the corresponding profile coefficient is reduced, the concrete piston is predicted to be in failure;
the fourth step: training and predicting data by utilizing LightGBM, XGboost and Catboost algorithms respectively, and finally fusing three prediction results by an averaging method;
in the fourth step, adding new features into a data set consisting of the features with original strong correlation to form a new training set; recording corresponding normal labels and fault labels in a label file of the training set, wherein the normal labels and the fault labels are respectively represented by 0 and 1; wherein 0 indicates that the corresponding piston of the corresponding sample has not failed during the future 2000-square concrete pumping task, and 1 indicates that the corresponding piston of the corresponding sample has failed during the future 2000-square concrete pumping task.
In the first step, the working condition data of the concrete pump truck are acquired by using the sensor and the edge end internet of things computing box, and the acquired working condition data comprise the working time of a piston (the accumulated working time after the piston is newly replaced, a numerical type), the engine speed, the oil pump speed, the pumping pressure, the hydraulic oil temperature, the flow gear, the distribution pressure, the discharge current amount (which is the corresponding working condition value and the numerical type of the pump truck), a low-voltage switch, a high-voltage switch, a stirring overpressure signal, a positive pump, a reverse pump (switching value) and the equipment type (the type and the category of the pump truck).
And in the second step, a KMeans function under a sklern.
When the data set consisting of the engine speed, the oil pump speed and the pumping pressure is subjected to trimerization, the number of clusters, namely the setting of a K value is determined according to multiple tests of the accuracy of prediction, the K value with the best prediction result is selected, and the parameter is set to be KMeans (K is 3).
In the third step, the contour coefficient is calculated for the clustering result by using silouette _ samples under the skearn. metrics package in python.
The above-described embodiment is only one specific embodiment of the present invention, and general changes and substitutions by those skilled in the art within the technical scope of the present invention are included in the protection scope of the present invention.