A kind of combination Clustering carries out the concrete piston failure prediction method of feature extraction
Technical field
The present invention relates to industrial equipment fault pre-alarming technical field, in particular to a kind of combination Clustering carries out feature and mentions
The concrete piston failure prediction method taken.
Background technique
Maintenance to production equipment, there are two main classes for traditional way, one is etc. failures occur after repair again, but this meeting
Lead to the halt production of unplanned property, economic loss is big;It is for second to be safeguarded with standing plans, but maintenance cost is high, when shutdown
Between it is long.Predictive maintenance, then by analysis malfunction history data and Real-time Monitoring Data, to the remaining life of equipment key part
Or failure carries out look-ahead early warning, and carries out maintenance and repair accordingly, to reduce equipment nonscheduled down time, reduce maintenance
Cost.
But there is no mature approaches of predictive maintenance at present.Traditional Knowledge based engineering method is mainly with related special
Based on the enlightening Heuristics of family and operator, connection relationship during qualitative or quantitative description between each unit,
Fault propagation mode etc., after abnormal sign occurs in equipment through the modes simulation process expert such as reasoning, deduction in monitoring
Inferential capability, to be automatically performed equipment fault early-warning and equipment monitoring.Such method is based on experience, can not establish accurate number
Learn model, it is usually poor with reliability from precision of prediction, it is difficult to according to existing status of equipment accurately early warning to future can
The equipment fault of energy.
A possibility that equipment fault early-warning and status monitoring are obtained according to equipment moving law or observation omen is true in equipment
Before just breaking down, the unusual condition of timely HERALD equipment takes appropriate measures, and can reduce equipment event to the greatest extent
Loss caused by barrier.As the scale and complexity of apparatus and engineering control system increasingly increase, to guarantee to produce
The safety and steady of journey, timely and effectively being monitored by reliable condition monitoring technology and diagnosing process exception just seems especially urgent
With it is important.
Concrete piston is the critical component and expendable parts of concrete mixer, and piston failure will lead to pump truck can not be just
Often work, while may cause other corollary equipments of entire building site can not normal construction, to bring sizable economic loss.
The specific operating condition of piston life and equipment etc. is closely related, and the real-time working condition data etc. of pump truck are uploaded to industry by Internet of Things
Internet cloud platform, the data based on accumulation establish suitable model, are expected to concrete piston during following certain task
The failure being inside likely to occur makes effective prediction and warning, so that operating personnel be reminded to carry out necessary maintenance before construction, keeps away
Exempt from bring economic loss due to UNPLANNED DOWNTIME.
Currently, there are following disadvantages for existing technology:
1. the non-machine learning algorithm of tradition can not predict concrete piston failure, once huge lose will be generated by breaking down
Damage, or even jeopardize worker's personal safety.
2. characteristic processing method is simply single in machine learning algorithm before, and potential information is excavated in harmless effective offer,
Strong feature is provided, forecasting accuracy is reduced.
Based on the above situation, the invention proposes the concrete piston failure predications that a kind of combination Clustering carries out feature extraction
Method.
Summary of the invention
In order to compensate for the shortcomings of the prior art, the present invention provides a kind of combination Clusterings being simple and efficient to carry out feature
The concrete piston failure prediction method of extraction.
The present invention is achieved through the following technical solutions:
A kind of combination Clustering carries out the concrete piston failure prediction method of feature extraction, which is characterized in that including following
Step:
Step 1: floor data is acquired using the industrial Internet of Things edge calculations box being mounted on concrete mixer, and
The floor data of acquisition is uploaded to cloud floor data library;
Step 2: after having acquired floor data, first between carrying out visual Pearson correlation analysis each feature,
The feature of strong correlation is selected, and three clusters are carried out to the feature for the strong correlation selected;
Step 3: calculating separately silhouette coefficient after obtaining cluster result to three central points of three clusters, obtaining three
Column new feature is used to describe the quality of cluster;
Step 4: XGBoost and CatBoost algorithm is trained and predicts to data respectively, most using Light GBM
It takes three prediction results to pass through the method averaged afterwards to be merged.
In the first step, the operating condition number of box acquisition concrete mixer is calculated using sensor and marginal end Internet of Things
According to floor data collected includes pistons work duration (newly changing cumulative activation duration after piston, numeric type), and engine turns
Speed, pump speed, pumping pressure, hydraulic oil temperature, flow gear, distribution pressure, row's magnitude of current (is the correspondence operating condition of pump truck
Value, numeric type), low tension switch, high-voltage switch gear stirs superpressure signal, and it is positive to pump, it is anti-to pump (switching value) and device type (pump truck
Type, classification type).
In the second step, after the floor data based on acquisition is analyzed, observation obtains engine speed, pump speed
And pumping pressure three two-by-two between all with 0.65 or more correlation, therefore by engine speed, pump speed and pumping pressure
Three feature floor datas of power establish new data set with these three features after copying, and utilize K-means clustering algorithm, right
Obtained new data set carries out three clusters.
In the second step, using the KMeans function under sklearn.cluster packet in python to engine speed,
The data set that pump speed and pumping pressure three are constituted carries out three clusters.
To engine speed, when the data set of pump speed and pumping pressure three composition carries out three clusters, of cluster
The accuracy test of many times of number, i.e. the basis of design prediction of K value determines that the K value for selecting prediction result best, parameter is set as
KMeans (K=3).
In the third step, using the silhouette_samples under sklearn.metrics packet in python to poly-
Class result calculates silhouette coefficient.
In the third step, under normal circumstances, the engine speed of concrete piston data, pump speed and three kinds of pumping pressure
The data variation of feature has more consistency, therefore Clustering Effect is more preferable, and silhouette coefficient is bigger;And the concrete to break down in the future
These three features of piston data will get muddled, and Clustering Effect will be deteriorated, and corresponding silhouette coefficient will become smaller.
In 4th step, new feature is added in the data set of the feature composition of original strong correlation, forms new instruction
Practice collection;And corresponding normal tag and faulty tag are recorded in the label file of training set, it is indicated respectively with 0,1;Wherein 0 table
Show that the sample corresponds to piston and do not break down during pumping 2000 side's concrete task in future, 1 indicates that the sample is corresponding
Piston future pump 2000 side's concrete task during failure has occurred.
The beneficial effects of the present invention are: this combines Clustering to carry out the concrete piston failure prediction method of feature extraction, tie
It closes pearson correlation property coefficient and K-means clustering method excavates the characteristic information for extracting relatively deep in feature, enhance
The accuracy and robustness of prediction greatly improve the reliability and accuracy of concrete piston fault pre-alarming, and can reduce need not
The production loss wanted.
Detailed description of the invention
Attached drawing 1 is the concrete piston failure prediction method schematic diagram that the present invention combines Clustering progress feature extraction.
Specific embodiment
In order to which technical problems, technical solutions and advantages to be solved are more clearly understood, tie below
Embodiment is closed, the present invention will be described in detail.It should be noted that specific embodiment described herein is only to explain
The present invention is not intended to limit the present invention.
This combines the concrete piston failure prediction method of Clustering progress feature extraction, comprising the following steps:
Step 1: floor data is acquired using the industrial Internet of Things edge calculations box being mounted on concrete mixer, and
The floor data of acquisition is uploaded to cloud floor data library;
Step 2: after having acquired floor data, first between carrying out visual Pearson correlation analysis each feature,
The feature of strong correlation is selected, and three clusters are carried out to the feature for the strong correlation selected;
Step 3: calculating separately silhouette coefficient after obtaining cluster result to three central points of three clusters, obtaining three
Column new feature is used to describe the quality of cluster;
Step 4: XGBoost and CatBoost algorithm is trained and predicts to data respectively, most using Light GBM
It takes three prediction results to pass through the method averaged afterwards to be merged.
In the first step, the operating condition number of box acquisition concrete mixer is calculated using sensor and marginal end Internet of Things
According to floor data collected includes pistons work duration (newly changing cumulative activation duration after piston, numeric type), and engine turns
Speed, pump speed, pumping pressure, hydraulic oil temperature, flow gear, distribution pressure, row's magnitude of current (is the correspondence operating condition of pump truck
Value, numeric type), low tension switch, high-voltage switch gear stirs superpressure signal, and it is positive to pump, it is anti-to pump (switching value) and device type (pump truck
Type, classification type).
In the second step, after the floor data based on acquisition is analyzed, observation obtains engine speed, pump speed
And pumping pressure three two-by-two between all with 0.65 or more correlation, therefore by engine speed, pump speed and pumping pressure
Three feature floor datas of power establish new data set with these three features after copying, and utilize K-means clustering algorithm, right
Obtained new data set carries out three clusters.
In the second step, using the KMeans function under sklearn.cluster packet in python to engine speed,
The data set that pump speed and pumping pressure three are constituted carries out three clusters.
To engine speed, when the data set of pump speed and pumping pressure three composition carries out three clusters, of cluster
The accuracy test of many times of number, i.e. the basis of design prediction of K value determines that the K value for selecting prediction result best, parameter is set as
KMeans (K=3).
In the third step, using the silhouette_samples under sklearn.metrics packet in python to poly-
Class result calculates silhouette coefficient.
In the third step, under normal circumstances, the engine speed of concrete piston data, pump speed and three kinds of pumping pressure
The data variation of feature has more consistency, therefore Clustering Effect is more preferable, and silhouette coefficient is bigger;And the concrete to break down in the future
These three features of piston data will get muddled, and Clustering Effect will be deteriorated, and corresponding silhouette coefficient will become smaller.
In 4th step, new feature is added in the data set of the feature composition of original strong correlation, forms new instruction
Practice collection;And corresponding normal tag and faulty tag are recorded in the label file of training set, it is indicated respectively with 0,1;Wherein 0 table
Show that the sample corresponds to piston and do not break down during pumping 2000 side's concrete task in future, 1 indicates that the sample is corresponding
Piston future pump 2000 side's concrete task during failure has occurred.
Embodiment described above, only one kind of the specific embodiment of the invention, those skilled in the art is in this hair
The usual variations and alternatives carried out in bright technical proposal scope should be all included within the scope of the present invention.