Lance A. Matheson

Sequencing mixed-model assembly lines with genetic algorithms

This research introduces the use of an artificial-intelligence based technique, genetic algorithms (GA), to solve mixed-model assembly-line sequencing problems. This paper shows how practitioners can comfortably implement this approach to solve practical problems. A substantial example is given for which GA produces a solution in just a matter of seconds that improves upon Toyotas Goal Chasing Algorithm. The new method is then investigated on a test bed of 80 problems. Results indicate GA generates an improved sequence over Goal Chasing on 50 of the problems and also shows a performance advantage of 2% across all 80 problems using Toyotas variability of parts consumption criterion. The paper concludes that further investigation to fine tune the GA methodology is warranted. It also points out that the GA approach can readily be used by practitioners to address a variety of managerial goals concurrently, such as inventory and work load equalization.

Improving software maintenance at Martin Marietta

Using data collected throughout a major project, the authors apply common statistical methods to quantitatively assess and evaluate improvements in a large contractors software-maintenance process. Results show where improvements are needed; examining the change in statistical results lets you quantitatively evaluate the effectiveness of the improvements. We selected a process-assessment methodology developed by J.E. Henry (1993) that follows Total Quality Management principles and is based on Watts Humphreys Process Maturity Framework. It lets you use a process modeling technique based on control-flow diagrams to define an organizations maintenance process. After collecting process and product data throughout the maintenance process, you analyze it using parametric and nonparametric statistical techniques. The statistical-analysis results and the process model help you assess and guide improvements in the organizations maintenance process. The method uses common statistical tests to quantify relationships among maintenance activities and process and product characteristics. The relationships, in turn, tell you more about the maintenance process and how requirements changes affect the product.<<ETX>>

Constructing optimal drug-testing plans using a Bayesian acceptance sampling model

Drug testing has become an accepted strategy for controlling drug use, particularly among individuals in the custody of the criminal justice system. Emphasis has been placed on testing those free in the community, either on pretrial release, probation, or parole. The drug-testing strategies applied to these populations-whom and how often to test-have evolved largely on an ad hoc basis. In this paper, we investigate optimal (cost-minimizing) drug-testing strategies as a means of achieving the efficient allocation of scarce resources to meet agency goals and objectives. We propose an analytic model based on individual decision theory and Bayesian acceptance sampling and apply the model to a hypothetical criminal justice population in which drug use is presumed to be highly prevalent.

ON THE USE OF BUFFER INVENTORIES TO MINIMIZE COSTS WITH AN UNRELIABLE MACHINE

In this paper, we discuss the problem of quality control with an unreliable machine which produces defects at a rate of Λ0, per unit when in-control and a rate of Lambda; 1, when out-of-control (where Λ1 Λ 0). Every h time periods, we sample n units, count the number of defects, and (using a process based on a Shewart c-chart) test the hypothesis that the machine is in control by comparing the total number of defects to an upper control limit (UCL). More important, we introduce the concept that a buffer inventory which immediately follows the unreliable machine may reduce expected total costs. This buffer serves to delay the movement of items from the unreliable machine to the next stage of the production process. In this way, we can isolate and repair most defective items before they are embedded in a product downstream or sold to customers where repair is more costly. To search for the optimal control policy, we find bounds for n, h, and UCL; given values for these variables, we show how the optimal buf...

Monitoring Drug Use Using Bayesian Acceptance Sampling: The Illinois Experiment

Bayesian acceptance sampling was used to monitor illegal drug use in a population of probationers. The study utilizes an economic model of drug testing based on single-sample, single-attribute acceptance sampling. This approach reduces from 100% the amount of testing which must be done to monitor the use of illegal drugs in the population and provides a decision rule, vis-i-vis a sampling plan, that specifies under what sampling outcome the entire population must be tested. The objective is to minimize the expected total cost of a drug testing program while ensuring that the proportion of users in the population is not increasing over time. A field study of the acceptance sampling approach was conducted using probationers assigned to Intensive Drug Supervision Programs in six Illinois counties. The degree to which drug testing results were reported to probation officers was controlled during the experiment. Counties were assigned to receive: no feedback of drug-test results; random proportion of feedback using Bayesian acceptance sampling plans; and 100% feedback-the status quo situation. Results show that those counties using acceptance sampling could have reduced the number of drug tests performed without increasing drug use. In those counties with no feedback, there was an upward trend over time in the proportion testing positive for drug use. Acceptance sampling-based drug testing programs are now being implemented by Illinois probation offices.

Quality control and social processes: a case for acceptance sampling

The “at the source” emphasis of total quality management (TQM) has reduced the reliance on post‐production statistical quality control approaches such as acceptance sampling. In cases where it is appropriate more proactive approaches such as statistical process control have improved productivity in manufacturing environments. For social processes where the inputs are ill‐defined and the outputs are difficult to measure, traditional quality control approaches have rarely been applied. Addresses the problem of monitoring use of illegal drugs, a critical social problem. Because the inputs, the use of drugs, are not easy to document and the process which results in an individual’s decision to use drugs is too complex to model, one must rely on the detection of drugs as a measurement of drug abuse. The behaviour of interest is the detection of illegal drug use through urine testing. The technique for monitoring this behaviour in a population of interest is single‐attribute, Bayesian acceptance sampling. Applies a partial drug‐testing methodology based on single‐attribute acceptance sampling to a population of probationers in Madison County, Illinois, USA. The approach offers probation offices with a lower cost approach to monitoring drug use among populations of known drug users. The use of acceptance sampling allows Madison County to reduce the total cost of testing by reducing the total amount of testing that must be done to monitor use of drugs among their probation populations.

Cost-effective drug testing in the transportation industry

Random testing for drugs and alcohol has become a critical issue for agencies and firms who employ ‘safety-sensitive’ transportation workers. The recent tightening of industry standards by the Depa...

On sequential versus random sampling in statistical process control

In statistical process control, a number of items are selected from all items produced every h time units (which we will refer to as an inspection period); these items are used to make inferences about the state of an unreliable machine or process. This paper considers an unreliable process which can shift from an acceptable in‐control state to an unacceptable out‐of‐control state. Based on a Shewhart‐type c‐chart, this paper extends the framework developed in Klastorin et al. to define the expected number of samples needed to confirm that the process shift has occurred when we use a sequential sample of the last n items produced in an inspection period. Comparing this result to the case where a random sample is used, we show that the probability of detecting the shift using a sequential sample is greater than or equal to the probability of detecting the shift using a random sample. Thus, sequential samples will result in a control chart that requires fewer expected samples to detect a shift and has lower expected total costs.

A multiple objective approach for constructing attribute-sampling plans for non-serial assembly items

A number of methods exist for determining fraction-defective sampling plans for single items. However, most methods currently exist for determining a set of optimal sampling plans for serial assembly processes. This paper develops a risk and cost-based model which explicitly considers the effect of the quality of lower level items on the quality of the final product. The model allows us to determine a minimum cost set of sampling plans for all items in a multi-level, non-serial product. We study the performance of the model with an algorithm for solving nonlinear, multiple-criteria formulations.

Statistical quality control and social processes: A drug testing application

Abstract Traditional acceptance sampling procedures have been used to monitor the quality of outgoing items from a production process within a manufacturing environment. Generally, however, this has represented a reactive, rather than a proactive, approach to quality control. Importantly, the manufacturing sector has refocused itself on more proactive techniques that attempt to improve item quality by repairing the process at the closest feasible point of intervention. This type of intervention is less possible, however, in most social service environments since: (1) the process may not be visible; and (2) the relationship between intervention and outcome is not well-understood and/ir well-defined. Largely because of the complexity of social services, statistical quality control procedures have not generally been applied to improve process quality or to otherwise affect process outcomes. Because of these difficulties, the benefits of statistical quality control procedures and their ability to make processes more efficient and/or effective have largely been ignored in the literature on quality control. However, provided one can identify an objective outcome measure from a process (regardless of the complexity of that process), and make some assumptions about the prior distribution, acceptance sampling can be an appropriate and useful technique for monitoring process outcomes. Indeed, for many social processes, acceptance sampling, or testing “after the fact” may be the only approach available for monitoring shifts in process quality. We here demonstrate the utility of Bayesian acceptance sampling in the context of a timely social process, testing a population for the use of illegal drugs. The use of drugs, and the desire on the part of businesses and the criminal justice system to control or deter use has been a major focal point for policy and decision makers for more than a decade. Given that budgets to institute drug testing and/or screening programs are not unlimited, a technique that reduces the cost of a drug treatment program while maintaining deterrence and monitoring effects would indeed be useful to practitioners. We thus propose an application within the framework of an economic model of drug use, and show that adoption of the testing approach can reduce the expected cost of testing.