Sandro Saitta

Data mining techniques for improving the reliability of system identification

A system identification methodology that makes use of data mining techniques to improve the reliability of identification is presented in this paper. An important aspect of the methodology is the generation of a population of candidate models. Indications of the reliability of system identification are obtained through an examination of the characteristics of the population. Data mining techniques bring out model characteristics that are important. The methodology has been applied to several engineering systems.

A Bounded Index for Cluster Validity

Clustering is one of the most well known types of unsupervised learning. Evaluating the quality of results and determining the number of clusters in data is an important issue. Most current validity indices only cover a subset of important aspects of clusters. Moreover, these indices are relevant only for data sets containing at least two clusters. In this paper, a new bounded index for cluster validity, called the score function (SF), is introduced. The score function is based on standard cluster properties. Several artificial and real-life data sets are used to evaluate the performance of the score function. The score function is tested against four existing validity indices. The index proposed in this paper is found to be always as good or better than these indices in the case of hyperspheroidal clusters. It is shown to work well on multi-dimensional data sets and is able to accommodate unique and sub-cluster cases.

Feature selection using stochastic search: An application to system identification

System identification using multiple-model strategies may involve thousands of models with several parameters. However, only a few models are close to the correct model. A key task involves finding which parameters are important for explaining candidate models. The application of feature selection to system identification is studied in this paper. A new feature selection algorithm is proposed. It is based on the wrapper approach and combines two algorithms. The search is performed using stochastic sampling and the classification uses a support vector machine strategy. This approach is found to be better than genetic algorithm-based strategies for feature selection on several benchmark data sets. Applied to system identification, the algorithm supports subsequent decision making.

Rational design of measurement systems using information science

Note: IABSE Report 92 CDROM Reference IMAC-CONF-2006-035 Record created on 2007-08-14, modified on 2016-08-08

Optimal Sensor Placement for Damage Detection: Role of Global Search

Optimal sensor placement is one that maximizes the likelihood of identifying future damage models. Based on assumptions from engineers, damage models of a structure are simulated and their predictions are computed. Computational approaches are used to place sensors at locations that maximize the chances of identifying damage. This paper studies the application of global search for optimal sensor placement. The global search methodology uses stochastic sampling to find optimal locations for sensors. In a previous study, Robert-Nicoud et al. proposed a greedy strategy that places sensors sequentially at locations where model predictions have maximum entropy. Performance of the two strategies are compared for the Schwandbach bridge in Switzerland. The results show that global search is better for designing measurement systems on a previously unmonitored structure while the greedy algorithm is better for incremental measurement- interpretation strategies.

Multiple-Model Updating to Improve Knowledge of Structural System Behavior

Note: CDROM Reference IMAC-CONF-2006-046 Record created on 2007-08-14, modified on 2016-08-08

Supporting engineers during system identification

Note: CDROM Reference IMAC-CONF-2005-035 Record created on 2007-08-14, modified on 2016-08-08

Combining two data mining methods for system identification

System identification is an abductive task which is affected by several kinds of modeling assumptions and measurement errors. Therefore, instead of optimizing values of parameters within one behavior model, system identification is supported by multi-model reasoning strategies. The objective of this work is to develop a data mining algorithm that combines principal component analysis and k-means to obtain better understandings of spaces of candidate models. One goal is to improve views of model-space topologies. The presence of clusters of models having the same characteristics, thereby defining model classes, is an example of useful topological information. Distance metrics add knowledge related to cluster dissimilarity. Engineers are thus better able to improve decision making for system identification and downstream tasks such as further measurement, preventative maintenance and structural replacement.

KnowPrice2: intelligent cost estimation for construction projects

Correct estimation of costs of construction projects is the key to project success. Although mostly established in early project phases with a rather limited set of project data, estimates have to be precise. In this paper, a methodology for improving the quality of estimates is presented in which data from past projects along with other knowledge sources are used. Advantages of this approach as well as challenges are discussed.