Ph'D' Research Proposal Leveraging Operational Data For Intelligent Decision Support in Construction

1
Ph.D. Research ProposalLeveraging Operational
Data For Intelligent Decision Support in
Construction Equipment Management

• Hongqin Fan
• Provisional Ph.D. Candidate in
• Construction Engineering and Management
• University of Alberta
• April 24, 2006

2
Agenda

• Introduction
• Problem Statement and research motivation
• Related research
• Research methodology
• Data warehousing for equipment management
• Resolution-based outlier mining algorithm
• AutoRegression Tree based Prediction
• AutoRegression Tree based time series forecasting
• Expected contributions
• Summary and conclusions

3
Introduction
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Definition of Construction Equipment Management
• Manage equipment resources to maximize return of
  capital investments and satisfy the needs of
  project management in a timely and cost-effective
  manner (Adapted according to Vorster and
  Livermore 1994).
• Major construction equipment management tasks
• Corporate level Equipment acquisition, finance,
  life cycle costing
• Operational level logistics, maintenance and
  repair.

4
Introduction
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Recent trends in construction equipment
  management
• Computerized construction equipment management
• Increasing automation in data collection and
  control
• Management of large fleet
• Equipment acquisition and service market
  diversification.

5
Introduction
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Most data are collected and stored in electronic
  format.
• Commonly cited problems with the current
  equipment data
• Data problem noisy data, fragmented, difficult
  to retrieve
• Underutilization of data assets, due to lack of
  advanced computer tools.
• This research will improve the current situation
  through data warehousing and data mining.

6
Introduction
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
The equipment management team has more advanced
tools for decision support in addition to the
summary reports.
7
Introduction
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• The research will achieve the following
  objectives
• Build a prototype construction equipment data
  warehouse as the enterprise data source for
  decision support. Explore the opportunities and
  challenges at different stages of data
  warehousing, including planning, design and
  implementation, for equipment management
• Design and test of a novel nonparametric outlier
  mining algorithm for generic problem detection in
  construction equipment data, as well as other
  engineering data.
• Testing, evaluation and modification of current
  data mining algorithms for decision support in
  construction equipment management
• Design and implement the prototype intelligent
  equipment management system using integrated
  equipment data warehouse and embedded data mining
  models make recommendations on planning and
  design of an intelligent system.

8
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• An equipment maintenance management system,
  called MTrack, was developed by NSERC/Alberta
  Construction Research Chair and used by Standard
  General Inc. (SGI) since 1997, generating a large
  data collection on equipment management.
• Based on the case of SGI and common to the
  industry, it is found that a number of problems
  undermine the usability of the collected data
• Equipment data
• Data quality issues
• Scattered data sources
• Most data are stored in relational database with
  high performance in data storage/updating but
  limited capability in data analysis
• Equipment data utilization
• Large amounts, but low rate of utilization for
  decision support in equipment management
• Lack of advanced computer tools for decision
  support using these data. Data analysis is
  commonly conducted by exporting data to
  statistical tools or spreadsheets.

9
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Relational databases are back-ends of most
  equipment management systems
• Equipment data stored in relational database are
  considered to be superior than these in
  applications, spreadsheets, or text files for
  analysis
• Still, equipment data in a relational database
  suffer two drawbacks in terms of data analysis
• Organized in relational data model, which is
  optimized for data adding/updating, but does not
  perform well in data retrieval
• Extracting data from equipment database can only
  be performed by database specialists, as a
  results, the user has limited control over what
  data can be extracted

10
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• This research is motivated by the recent
  development in data warehousing and data mining
  techniques
• Affordable as commercial products
• Integration with other applications through
  standard communication protocols
• And their capability of improving data quality,
  data structure and knowledge extraction in terms
  of data utilization.

11
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Data warehousing for Construction Equipment
  Management
• A warehouse is a subject-oriented, integrated,
  time-variant and non-volatile collection of data
  in support of management's decision making
  process Inmon 1996.
• Construction equipment data warehouse improves
  data quality by preprocessing, and integrate data
  by pulling needed data from various sources.
• Construction equipment data warehouse uses
  multidimensional data model and is built around
  the subjects of equipment management, making it
  possible to perform interactive data analysis by
  end users.
• Construction equipment data warehouse can serve
  as significantly improved data source for
  knowledge discovery and prediction.

12
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Data mining for identification of patterns,
  making prediction and forecasting
• Data mining is the process of extracting non
  trivial, implicit, previously unknown and
  potentially useful information from large
  collections of data Frawley et al. 1992
• Data mining can
• Identify patterns (common patters or unusual
  patterns)
• Create data mining models for exploration of
  hidden knowledge in data
• Use model for prediction and forecasting.

These tasks are common problems in decision
making for equipment management.
13
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Data mining models are data-driven, which means
  the data mining algorithms derive models from
  historical data, and can update the model
  dynamically if there are any changes in data.
• Comparison with Statistical, mathematical and
  simulation models

14
Problem Statement and Research Motivation
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Three research topics on data mining are selected
15
Related Research
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Researches on construction equipment management
  are primarily focused on
• Automation and robotic technologies
• Real-time data communications and information
  processing
• Statistical and analysis modeling for decision
  support in equipment management.
• Researches in intelligent equipment/assets
  management systems are conducted in the
  maintenance operations of the following
  facilities
• Power plants,
• Industrial plants,
• Military facilities

16
Related Research
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Researches on applying data warehousing
  technology in following areas of construction
  industry
• Inventory management of construction materials
  Chau et al. 2004
• Document management for multi-party and
  multi-purposes Ma et al. 2005
• Data mining techniques are applied in various
  researches and applications of construction
  industry
• Estimation of construction productivity using
  artificial neural network Lu et al. 2002
• Construction delay evaluation using C 4.5
  decision tree Soibelman and Kim 2002
• Classify and quantify cumulative impact of change
  orders on productivity using decision tree Lee
  et al. 2004
• Automatic classification of construction document
  using Support Vector Machine (SVM) Caldas et al.
  2004
• Preliminary research is conducted on data
  preprocessing and knowledge discovery from
  construction data Soibelman and Kim 2002

17
Research Methodology
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• The research is based on the case of equipment
  management in Standard General Inc., Alberta,
  Canada.
• An overview of research scope

18
(No Transcript)
19
Research Methodology (Data warehousing for
equipment management)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Procedures for construction equipment data
  warehousing
• Identify all the data sources containing related
  data for equipment management
• Extract, transform and load data into the data
  warehouse repository
• Expose data for interactive data analysis,
  knowledge discovery or reports.

20
Research Methodology (Data warehousing for
equipment management)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting
Procedures for building equipment data warehouse
21
Research Methodology (Data warehousing for
equipment management)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Data modeling and architectural design
• Design of multidimensional data model for each
  subject, using star schema
• Each data model contains a fact table surrounded
  by a number of dimension tables
• Descriptive attributes are usually formed in
  hierarchy to allow for data analysis at different
  levels of details.
• Many data cubes share dimensions in equipment
  data warehouse, use Bus Matrix model, proposed by
  Kimball and Ross2002, for architectural design

22
Research Methodology (Data warehousing for
equipment management)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting
Measurements of performance in equipment
management
Dimensions describing the measurements in fact
table
Multidimensional data model for subject Repair
Cost
23
Research Methodology(Resolution-based outlier
mining algorithm)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Outlier mining identifies irregular patterns, or
  inconsistent records from a dataset.
• Current statistical methods (e.g. multivariate
  outlier detection) and outlier mining algorithms
  do not perform well in engineering applications.
• An non-parametric outlier mining algorithm based
  on resolution change is proposed in this
  research.

24
Research Methodology(Resolution-based outlier
mining algorithm)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Why is it possible to identify outliers during
  resolution change?
• Given a set of data objects, the underlying
  clusters and outliers change when increasing
  or decreasing the resolution of data objects.
• This makes it possible to identify outliers by
  consecutively changing the resolution of a set of
  data objects and collect pre-defined statistical
  properties.
• A resolution based outlier factor (ROF) is
  defined for measuring the degree of outlying of a
  data point.
• Overall procedures for the algorithm

25
Flowchart for resolution-based outlier mining
algorithm
26
Research Methodology(Resolution-based outlier
mining algorithm)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting
Preliminary experimental results on a synthetic
dataset
Top-10 outliers detected from a synthetic
200-tuple Dataset
27
Research Methodology(Resolution-based outlier
mining algorithm)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• The research will compare the proposed algorithm
  with the current distance-based outlier mining
  algorithm Knnor and Ng 1998 and Local Density
  Based outlier mining algorithm Breunig et al.
  2000 from the following perspectives
• Results of experimental tests on synthetic
  datasets and real life equipment datasets
• Explain the test results from the outlier
  definition and outlier mining algorithms
• Compare the pros and cons of the three algorithms
  in engineering applications.

28
Research Methodology(AutoRegression Tree based
prediction)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• For numerical target attribute, prediction
  problem in data mining is to estimate the target
  attribute based on a set of known attributes
  (categorical or numerical values).
• AutoRegression Tree (ART) data mining algorithm
  Meek et al. 2002 is a hybrid algorithm of
  decision tree and multivariate linear regression,
  designed for prediction purpose.
• Using the training dataset, ART algorithm grows a
  top-down decision tree with a linear regression
  model in each leaf node.

29
Research Methodology(AutoRegression Tree based
prediction)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Information gain is used to select attributes and
  splits for C4.5 decision tree growing Kantardzic
  2003.

30
Research Methodology(AutoRegression Tree based
prediction)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Least square method is used to build the
  multivariate linear regression model in each leaf
  node.
• An example of ART estimation is to evaluate work
  orders estimated by the equipment superintendent
• Given the factors of impact, such as equipment
  manufacturer, age, component, repair type,
  estimated hours etc, what is the likely error of
  estimates?

31
Research Methodology(AutoRegression Tree based
prediction)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting
Part of the Induced Work Order Evaluation ART
Model
32
Research Methodology(AutoRegression Tree based
prediction)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• The research will evaluate the ART algorithm in
• Estimate accuracy
• Interpretability of the derived model
• Flexibility in solving different problems in
  equipment management.
• And compare with Categorization and Regression
  Tree (CART) Breiman et al. 1984.

33
Research Methodology(AutoRegression Tree based
time series forecasting)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Time series data is a series of data collected
  over successive increments of time
• Time series forecasting predicts future values of
  a time series, based on historical observations
  and assuming the current trend continues.

34
Research Methodology(AutoRegression Tree based
time series forecasting)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Traditional statistical methods decompose a time
  series into four basic movements trend, cyclic,
  seasonal, and irregular movements.
• The basic assumption of time series forecasting
  model is AutoRegression, which assumes the
  current value of a time series depends on its
  previous n observed values.

Noise

• In its most simple case, use linear regression
  and solve the problem using least square method.

35
Research Methodology(AutoRegression Tree based
time series forecasting)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Current approaches for solution
• Statistical modeling and forecasting
• ARMA (AutoRegression Moving Average) is a
  representative statistical approach for modeling
  and forecasting.
• Neural Network forecasting
• Use Neural Network (NN) to replace the regression
  model.
• This research use ART model for forecasting
• Use AutoRegression Tree (ART) data mining model
  to replace the regression model.

36
Preliminary test results
Forecasting results using ART model for
Predicting monthly equipment repair and
maintenance costs in Standard General Inc.
37
Research Methodology(AutoRegression Tree based
time series forecasting)
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions
Equip. data warehousing
RB-outlier mining
ART prediction
ART forecasting

• Time-series prediction based on ART data mining
  algorithm will be compared with ARMA Statistical
  method and neural network from the following
  perspectives
• Accuracy of prediction
• Extensibility and transparency of prediction
  model
• Pros and cons in system integration for equipment
  management.

38
Expected Contributions
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• This research will provide guidelines for
  applying data warehousing technology to
  construction equipment management for improved
  decision support. These include the
  opportunities, challenges, and suggestions for
  planning and design of an equipment data
  warehouse
• A novel non-parametric outlier mining algorithm
  is proposed for generic problem detection in both
  equipment management and other engineering
  applications. This will contribute to the body of
  knowledge in data mining community.
• Current data mining algorithms, such as
  AutoRegression Tree, will be tested, evaluated
  and modified for intelligent decision support in
  construction equipment management. This research
  will report my findings and make recommendations
  on the general application of data mining
  technology in construction equipment management.
• This research will summarize and make
  recommendations on the architectural design and
  implementation of an intelligent equipment
  management information system using combined data
  warehousing/data mining techniques, to meet
  industrial expectations.

39
Conclusions
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• This research will address data management issues
  and facilitate transformation of data into useful
  information and knowledge which the equipment
  management team can act upon.
• High level design of intelligent systems for
  equipment management, and detailed study on a
  novel data mining algorithm, evaluation of
  current ART data mining algorithms for
  engineering applications are the focuses of this
  research
• This research addresses both academic issues and
  real life application issues, therefore it will
  directly benefit the construction industry.

40
References
Introduction
Research Motivation
Related Research
Research Methodology
Expected Contributions
Conclusions

• Breiman, L., Friedman, J., Olshen, R. and Stone,
  C. (1984). Classification and Regression Trees,
  Chapman Hall/CRC Press, Boca Raton, FL.
• Breunig, M., Kriegel, H., Ng, R., and Sander, J.
  2000. LOF identifying density-based local
  outliers, Proceedings of ACM SIGMOD 2000
  International Conference on Management of Data,
  Dalles, TX, USA
• Caldas, C. H., Soibelman, L. and Han J. (2002)
  Automated Classification of Construction Project
  Documents. ASCE Journal of Computing in Civil
  Engineering, 16(4), 234-243
• Chau, K.W., Cao, Y., Anson, M., and Zhang J.
  2002. Application of Data Warehouse and Decision
  Support System in Construction Management.
  Automation in Construction, 12 213224.
• Frawley, w., Piatetsky-Shapiro, G. and Matheus,
  C. (1992). Knowledge Discovery in Databases An
  Overview. AI Magazine, Fall 1992, pp. 213-228.
• Inmon, W.H. (1996). Building the Data Warehouse.
  John Wiley Sons, New York.
• Kantardzic, M. (2003). Data Mining Concepts,
  Models, Methods, and Algorithms. John Wiley
  Sons, Inc. NJ. USA.
• Kimball, R. and Ross, M. 2002. The Data Warehouse
  Toolkit The Complete Guide to Dimensional
  Modeling, second edition, John Wiley Sons,
  Inc., New York, pp. 1388.
• Knorr, E., and Ng, R. (1998) Algorithms for
  mining distance-based outliers in large
  datasets. Proceedings of Very Large Data Bases
  Conference, New York, USA
• Lee, M., Hanna, A.S. and Loh, W.Y. (2004).
  Decision Tree Approach to Classify and Quantify
  cumulative Impact of Change Orders on
  Productivity. J. Comp. in Civ. Engrg., ASCE,
  18(2), 132
• Lu, M., AbouRizk, S.M. and Hermann U.H. (2002).
  Estimating labor productivity using probability
  inference neural network J. Comp. in Civ.
  Engrg., ASCE, 14(4), 241-248
• Ma, Z., Wond, K.D., Heng, L. and Jun Y. (2005)
  Utilizing exchanged documents in construction
  projects for decision support based on data
  warehousing technique. Automation in
  Construction, 14(3), 405-412
• Meek, C., Chickering, D.M. and Heckerman, D.
  (2002). Autoregressive Tree Models for
  Time-Series Analysis. Proceedings of the 2nd
  SIAM International Conference on Data Mining,
  Arlington, VA, USA.
• Soibelman, L and Kim, H. (2002). Data
  Preparation Process for Construction Knowledge
  Generation through Knowledge Discovery in
  Databases. ASCE Journal of Computing in Civil
  Engineering, 16(1), 39-48
• Vorster M. C. and Livermore M. E. (1994).
  Executive development for equipment managers.
  Proceedings of conference Equipment Resource
  Management into the 21st Century. Nashville,
  Tennessee, pp 87 95of conference Equipment
  Resource Management into the 21st Century.
  Nashville, Tennessee, pp 87 95

41
H. Fan, Prov. Ph.D. Candidate
University of Alberta