Norma Faris Hubele

Advancement in the application of neural networks for short-term load forecasting

An improved neural network approach to produce short-term electric load forecasts is proposed. A strategy suitable for selecting the training cases for the neural network is presented. This strategy has the advantage of circumventing the problem of holidays and drastic changes in weather patterns, which make the most recent observations unlikely candidates for training the network. In addition, an improved neural network algorithm is proposed. This algorithm includes a combination of linear and nonlinear terms which map past load and temperature inputs to the load forecast output. The search strategy and algorithm demonstrate improved accuracy over other methods when tested using two years of utility data. In addition to reporting the summary statistics of average and standard deviation of absolute percentage error, an alternate method using a cumulative distribution plot for presenting load forecasting results is demonstrated. >

Back-propagation pattern recognizers for X¯ control charts: methodology and performance

Abstract The pattern recognition algorithm presented here is based on the perception that, as automated data collection becomes more widespread in manufacturing processes, the monitoring of control charts will be performed by computer-based algorithms. These algorithms will have to detect unnatural patterns to assist in the correction of assignable causes. The work currently being performed in addressing the application of pattern recognition to control charts is directed toward answering this need. In this paper, a control chart pattern recognition methodology based on the back-propagation algorithm, a neural computing theory, is presented. This classification algorithm, suitable for real-time statistical process control, evaluates observations routinely collected for control charting to determine whether a pattern, such as a trend or cycle, exists in the data. The foundation of the algorithm is based on the neural network concepts of constructing and training a network in the types of patterns to be detected. These concepts mimic the trained operators ability to detect patterns. Here, the pattern recognizer is trained and tested with the consideration of Type I error (finding a pattern where none existed) as well as Type II error (failing to detect a known pattern). Performance measures sensitive to these types of errors are used to evaluate the algorithms performance on an extensive series of simulated patterns of control chart data. This approach is promising because of its flexible training and high-speed computation.

Identification of seasonal short-term load forecasting models using statistical decision functions

A hierarchical classification algorithm is applied to hourly temperature readings to divide the historical database into seasonal subsets. These subsets are used to statistically identify and fit a response function for each season. These functional models constitute a library of models useful to the power scheduler. For a particular day, the appropriate model is selected by performing discriminant analysis. This approach is illustrated using data from a summer peaking utility. This application demonstrates that an entire procedure for specifying forecasting models can be formed with currently available statistical software. Furthermore, the models can be implemented on a microcomputer spreadsheet. >

Design of a knowledge-based expert system for statistical process control

Abstract A comprehensive expert system design is presented for all problem-solving aspects of statistical process control. Issues concerning the integration of monitoring, interpreting, diagnosing, planning, and statistical consulting for statistical process control are addressed. All aspects of the expert system are discussed. In addition, a new similarity measure useful in a clustering approach to the knowledge organization is proposed. An example session is presented which diagnoses a cyclic pattern in a surface-finishing operation.

Using Confidence Intervals to Compare Process Capability Indices

Process capability indices are widely used to measure process performance. In situations such as selecting a supplier and assessing process improvement, it is of interest to compare capability indices for two different processes or the same process before and after an adjustment. In this paper, we consider several methods for performing this comparison on the indices Cpk and Cpm . The methods are compared using a computer simulation. Recommendations are provided for selecting an appropriate method based on power and test size computations.

C pk index estimation using data transformation

Abstract Process capability indices (PCIs) are used in industry to assess percentages of nonconforming parts. An underlying assumption is that the output process measurements are distributed as normal random variables. When normal distributions are assumed, but different distributions are present - such as skew, heavy-tailed, and short-tailed distributions - the percentages of nonconforming parts are significantly different than the computed PCIs indicate. Data arising from nonnormal distributions can sometimes be transformed to conform to the normality assumption and the PCIs computed for the transformed data. In this paper, the effect of the transformation on the estimate of nonconforming parts is examined for three examples of nonnormal distributions - gamma, lognormal, and Weibull. The results of this experimental analysis suggest that data transformation can be useful for estimating an interval for C pk values and the number of nonconforming parts.

QUANTILES OF THE SAMPLING DISTRIBLITION OF C

Point estimates of the capability index Cpm are generated from sample observations from manufacturing processes and are used to make comparisons of processes. Decisions based on these quantities should necessarily include information about the variabili..

Using Experimental Design to Assess the Capability of a System

Experimental design techniques were applied to the task of characterizing the inspection capability of the machine vision component of an automated laser hole-drilling and inspection system for gas turbine engine manufacturing. The system uses a laser t..

Boltzmann machines that learn to recognize patterns on control charts

Boltzmann machines (BM), a type of neural networking algorithm, have been proven to be useful in pattern recognition. Patterns on quality control charts have long been recognized as providing useful information for correcting process performance problems. In computer-integrated manufacturing environments, where the control charts are monitored by computer algorithms, the potential for using pattern-recognition algorithms is considerable. The main purpose of this paper is to formulate a Boltzmann machine pattern recognizer (BMPR) and demonstrate its utility in control chart pattern recognition. It is not the intent of this paper to make comparisons between existing related algorithms. A factorial design of experiments was conducted to study the effects of numerous factors on the convergence behavior and performance of these BMPRs. These factors include the number of hidden nodes used in the network and the annealing schedule. Simulations indicate that the temperature level of the annealing schedule significantly affects the convergence behavior of the training process and that, to achieve a balanced performance of these BMPRs, a medium to high level of annealing temperatures is recommended. Numerical results for cyclical and stratification patterns illustrate that the classification capability of these BMPRs is quite powerful.

QUALITY EVALUATION USING GEOMETRIC DISTANCE APPROACH

Quantitatively assessing quality and using this assessment for competitive benchmarking and diagnostics of manufactured part failure are very important for continuous improvement in modern manufacturing industries. Process capability analysis often entails characterizing or assessing process specification or quality characteristics. When these quality characteristics are related, the analysis should be based on a multivariate statistical technique. A current problem in multivariate quality control is that there is no consensus about a methodology for assessing capability. Thus, the critical first step in instituting a multivariate control scheme is not well defined. While numerous authors have recently proposed alternative definitions of multivariate capability indices, those methods may not be practical in some cases. In this research, a new process control variable, geometric distance (GD), for assessing or evaluating the quality of manufactured product is developed and investigated for reducing dimensionality. The theoretical distribution of the geometric distance is investigated and a suitable performance metric of the multivariate process data is proposed. Finally, some real data are used to demonstrate the capability of the proposed method.