Mine Planning and Equipment Selection 2002

Modeling condition and performance of mining equipment

Tad S. Golosinski and Hui Hu

Department of Mining Engineering, University of Missouri-Rolla, Rolla, MO 65409-0450, USA

ABSTRACT: The paper follows on the earlier MPES paper that suggested use of data mining techniques to model operation of mining equipment. It reports on the new developments that concentrated on modeling of performance and condition of mining trucks based on the analysis of digital data collected in the field by truck vital sign information and management system. The modeling was done using several algorithms as provided in the Intelligent Miner software package of IBM. The paper presents approaches with the models built using decision tree algorithm. The models developed as the result of this work allow for projection of truck condition and performance into future with reasonably high accuracy. As such they allow for better control of mining operation and are expected to find numerous applications in mines worldwide.

Mine Planning and Equipment Selection 2002

1 INTRODUCTION

Modern mining equipment equipped with numerous sensors monitors its condition and performance. Data collected by these sensors is used to alert the operator to existence of abnormal operating conditions and to perform emergency shutdown if the pre-set upper or lower limits of the monitoring parameters are reached. This data is also used for post-failure diagnostics and for reporting and analysis of equipment performance.

It is believed that availability of this voluminous data, together with availability of sophisticated data processing methods and tools, may allow for extraction of additional information contained in the data. One method that may permit this is data mining (Golosinski, T. S. 2001) (Golosinski, T. S. and Hu, H. 2001).

The research presented in this paper investigates use of the data collected from various sensors installed on the mining truck for construction of a truck model, which may allow for reliable prediction of both the truck performance and its condition into the future. Subject to research was data collected by a variety of sensors installed on off-highway mining trucks that together constitute the VIMS (Vital Information Monitoring System) system of Caterpillar (Caterpillar 1999). The data mining tool was the IBM Intelligent Miner for Data (IBM 2000).

2 DATA DESCRIPTION

The data used in this research consists of snapshot (event recorder) and datalogger records, each containing values of 70 truck parameters measured over a period of time. The data was collected from 6 Caterpillar 789B trucks during their operation in a surface mine.

The snapshot stores a segment of truck history that contains values of all 70 monitored parameters recorded during the period of six minutes, each parameter value recorded once per second. The snapshot recording is triggered by one of a set of predefined events, usually occurrence of an abnormal situation where a specific parameter reaches a critical value. A snapshot record describes truck conditions from five minutes before the event to one minute after the event (Caterpillar 1999). In this paper, every snapshot record is called “event” for simplicity.

Unlike snapshot, the data logger records values of all truck parameters that are monitored by VIMS over varying periods of time, also at one-second intervals (Caterpillar 1999). The recording and its end are triggered manually, with individual records covering periods of up to 30 minutes of truck operation. Datalogger records do not have to be associated with any events.

Of the 70 truck parameters used in this research, values of 26 were recorded as categorical and the remaining 44 as numeric values. The examples of basic statistical description of both the categorical and numerical parameter values are presented in Table 1 and 2. Previously research investigated the data mining directly on the second data (Hu, H. and Golosinski, T. S. 2002). The new approach uses the statistical data calculated from one to three minute intervals. The major statistical parameters include:

Table 1. Example of categorical parameter values

Parameter Name / Modal Value / Modal Frequency (%)
ACTUAL_GEAR_352 / Neutral / 41.55
AFTRCLR_LVL_137 / OK / 98.95
BODY_LVR_727 / Not Moving / 95.4
BODY_POS_726 / Down / 93.7

Table 2. Example of numerical parameter values

Parameter Name / Minimum
Value / Maximum
Value / Mean
Value / Standard
Deviation
AFTRCLR_TEMP_110 / 0 / 95 / 41.8 / 12.8
AMB_AIR_TEMP_791 / 0 / 38.5 / 21.9 / 7.0
ATMOS_PRES_790 / 0 / 93 / 89.4 / 9.1
BOOST_PRES_105 / 0 / 164 / 31.0 / 50.1

Figure 1 explains how the model predicts one of the VIMS events, “high engine speed”. In this illustration, the VIMS data is aggregated into statistical data of every three minutes and “high engine speed” is the predicted VIMS event. The predicted label is the name of the class defined in Intelligent Miner for classification. If one three-minute data is classified as the “Eng_1” by the model, it recognizes this data has similar statistical characteristics with the first three minutes of the “high engine speed” snapshot. Because the event actually takes place at the end of the fifth minute of the snapshot (total of six minutes), it is possible that high engine speed will take place after two minutes according to the interpretation.

Figure 1. VIMS Event Prediction

As described in Figure 1, the one-minute model can provide 1-4 minutes early prediction; and the two-minute model can provide 1 and 3 minutes early prediction; and the three-minute model can only provide two minutes early prediction.

3 modeling DESIGN

3.1. Objective

Modeling was designed to evaluate and quantify the pattern of changes in parameter values as associated with various events. As the sensors installed on the truck activate the snapshot recorder when the predefined limit of a parameter is reached, the objective was to identify any patterns in parameter values that may allow for early failure recognition. These patterns were then used for prediction of future events by building a decision tree classification model of the truck. The model was to predict an occurrence of a selected event based on the pattern of changes in values of other parameters.

The “high engine speed” events recorded most frequently in the available VIMS data set was selected as the main targets of analysis. In addition other events, the data collected during normal operation was selected as class “other” as well. The “engine speed” is defined as the actual rotational speed of the crankshaft. For the modeled truck this event is activated when the engine speed reaches 2250 rpm and deactivated when the speed drops to 1900 rpm.

3.2. Data Mining Tools

IBM Intelligent Miner software package was used as the data mining tool. The basic algorithm used was SPRINT, a modified CART (Classification and Regression Tree). It was chosen in preference to the neural network classification algorithm as the Decision Tree approach is easier to interpret and understand by engineers, thus facilitating easy analysis of the truck failure pattern (IBM 1999).

The workings of SPRINT are similar to that of most popular decision tree algorithms, such as C4.5 (Quinlan J.R. 1993); the major distinction is that SPRINT induces strictly binary trees and uses re-sampling techniques for error estimation and tree pruning, while C4.5 partitions according to attribute values (Jang, J. and Sun, C. 1997). The GINI index is used to measure the misclassification for the point split by SPRINT algorithm. For a data set S containing examples from n classes, the gini(s) is defined as shown in Eq.(1) where pj is the relative frequency of class j in S. If a split divides s into two subsets s1 and s2, the index of the divided data ginisplit(s) is given by Eq.(2). The advantage of this approach is that the index calculation requires only the knowledge of distribution of the class values in each of the partitions (Breiman, L. and Friedman, J. 1984).

(1)

(2)

The tree accuracy is estimated by testing the classifier on the subsequent cases whose correct classification has been observed (Quinlan J.R. 1993). The v-fold cross-validation technique estimates the tree error rate. This estimation of error rate is used to prune the tree and choose the best classifier. More detail about this algorithm can be found elsewhere (Shafer, J. 1996).

3.3. Modeling Procedures

The two main procedures of data mining are training called also model construction, and testing called also model validation. In training mode, the function builds a model based on the selected input data. This model is later used as a classifier. In test mode, the function uses a set of data to verify that the model created in the training mode produces results with satisfactory precision.

In this work all available data was split into two parts. Bulk of the data, 86.4%, was used for model training. The remainder, 13.6% of available data, was used for model testing. The test data includes dataset #1 (random selection) and dataset #2 (whole snapshot and datalogger).

After three models are built on the one, two, and three minute statistical data, the error rate was defined and used for evaluating the performance of the training and the testing processes.

4 RESULTS AND DISCUSSIONS

The model built on three-minute statistical data has less than 5% training error rate, 19% error rate on test #1, and 14% error rate on test #2. The model shows better performance on unseen VIMS event prediction than one- and two- minute models (Table 3). However, the tradeoff is that this model can only provide two-minute early prediction with three classes, “Eng1”, “Eng2” and “Other”.

Table 3. Model Performance Comparison

Representative three-minute model output is shown in Figure 1-3 for training, test #1 and test #2 respectively as the confusion matrix. A confusion matrix for the pruned tree shows the distribution of the misclassifications. In every matrix, the number on the diagonal is the correct classification; others are the number of misclassification.

Figure 2. Training with Three-Minute Statistical Data

Figure 3. Test#1 with Three-Minute Statistical Data

Figure 4. Test#2 with Three-Minute Statistical Data

4.1. Model Performance Analysis

After the VIMS data is aggregated into statistical data at minute interval, the number of rows is dramatically reduced and the data mining process is much quicker than mining second data. Error rates on test data (unseen events) are reduced into 19% and 14% (Figure 2, 3) for test #1 and test #2 respectively, which means the model is more robust than using statistical data at one and two minute interval.

Figure 5. Average Error Rate Comparison

Figure 6. Error Rate Standard Deviation Comparison

In addition, the error rate, as well as the related error rate mean and standard deviation, is calculated on every class. As Figure 5 and 6, the three-minute model presents the best prediction performance in terms of the 3% and 21% average error rates and 3% and 26% standard deviation for training and tests.

Table 4. Error Rate Statistical Calculation

The three-minute model has only 3% and 21% average prediction error rate, but the high error rate standard deviation on tests makes the prediction unstable. The class “Other” is defined as the operation with normal engine speed. In three-minute model, 14% error rate of “Other” (Table 4) on tests statistically implies the 14% probability of the false alarm of the high engine speed. The class “Eng1” is defined as the two minutes before high engine speed and the 50% error rate (Table 4) might implies the 50% probability of the missing alarm about the high engine speed. Because the class “Eng2” is defined as a period overlapping with the snapshot triggered time, it doesn’t provide any prediction before the event. The model still has high probability of missing alarm.

4.2. Three-Minute Decision Tree Classification Model

The approach shows more knowledgeable decision tree (simpler) that may be presented as a binary decision tree (figure 7). Each interior node of the binary decision tree tests an attribute of a record. If the attribute value satisfies the test, the record is sent down the left branch of the node. If the attribute value does not meet the requirements, the record is sent down the right branch of the node. Three classes are marked with different colors at upper left corner. The solid circles are the decision nodes. The binary decision tree consists of the root node on top, followed by non-leaf nodes and leaf nodes. Branches connect a node to two other nodes. Root and non-leaf nodes are represented as pie charts. Leaf nodes are represented as rectangles.

Figure 7. Decision Tree Structure on Three-Minute Statistical Data

From this decision tree (as Figure 7), the root node, named “ENG_SPD_MAX”, classifies nearly all “Eng2” events into right leaf (69 out of 70) displaying as the yellow rectangle. The reason is that the “high engine speed” event is activated and recorded when the engine speed reaches the predefined limit at the second three-minute of the snapshot. The rule for this classification is:

If (ENG_SPD_MAX>=2184.25)

then class=ENG

The activation value of “high engine speed” event defined by VIMS is 2250 rpm. This difference with the value discovered by the decision tree model makes four events, which are not “Eng2” events, misclassified into “Eng2”.

The rest of events are further classified according to more complex rules. One of major rules classifies the “Eng1” event as follows (circled leaf in Figure 7):

If (ENG_SPD_MAX<2184.25)

and (TRBO_IN_PRES_MIN<87.75)

and (ENG_SPD_REGR_SYY<44991960)

and (RTR_LTR_SUSPCYL_REGR_INTERCEPT<-1688.9)

and (GEAR_SELECT_RANGE>=100.5)

then class=ENG1

This rule classified 14 “Eng1” events correctly and only one is misclassification. It uncovers the three-minute statistical characteristics of the two minutes before high engine speed.

5 CONCLUSIONs

This approach compresses the information into statistical table and provides the prediction with certain accuracy. It also gives the possibility to predict the event for two minutes earlier. However, the prediction accuracy needs further improvement and results need verified by more test data. The possible approach to improve the prediction accuracy is to add more statistical parameters and use more VIMS data.

REFERENCE

Breiman, L. and Friedman, J. 1984, Classification and regression tree . Wadsworth International Group.

Caterpillar, Inc. 1999, Vital information management System (VIMS): system operation testing and adjusting. Company publication.

Golosinski, T. S. 2001, data mining uses in mining. Proceedings, Computer Applications in the Minerals Industries (APCOM), Beijing, China, 2001, pp. 763-766.

Golosinski, T. S. and Hu, H 2001, Intelligent Miner-data mining application for modeling VIMS condition monitoring data. Proceedings of the Artificial Neural Networks in Engineering Conference (ANNIE 2001), St. Louis, Missouri, U.S.A.

Hu, H. and Golosinski, T. S. 2002, Failure pattern recognition of a mining truck with a decision tree algorithm. Mineral Resources Engineering (in print).

IBM (International Business Machines Corporation) 2000, Manual: “Using the Intelligent Miner for Data”. Company publication.

IBM (International Business Machines Corporation) 1999, Intelligent Miner for Data: enhance your business intelligence. Company publication.

Jang, J. and Sun, C. 1997, Neuro-fuzzy and soft computing. Prentice-Hall, Inc.

Quinlan, J. R. 1993, C4.5: Programs for machine learning. Morgan Kaufmann Publishers, Inc.

Shafer, J. 1996, SPRINT: a scalable parallel classifier for data mining. Proceedings of the 22nd VLDB Conference, Mumbai (Bombay), India, 1996.