Adaptive Control Approach for Software Quality Improvement

The following sections are included:

• PREFACE

• CONTENTS

This chapter proposes a graph-model-based approach to prioritizing the test process. Tests are ranked according to their preference degrees which are determined indirectly, i.e., through classifying the events. For construction of the groups of events, an unsupervised neural network is trained by adaptive competitive learning algorithm. A casestudy demonstrates and validates the approach.

Biologically inspired soft computing paradigms such as neural networks are popular learning models adopted in online adaptive systems for their ability to cope with the demands of a changing environment. However, the acceptance of adaptive controllers is limited by the fact that methods and tools for the analysis and verification of such systems are still in their infancy. Generic Verification and Validation (V&V) procedures do not exist. The reliability of learning, performance, convergence or prediction for neural network models and their applications is hard to guarantee.

In this paper, we present several statistical evaluation methods proposed for the V&V of neuro-adaptive systems. These methods include support vector data description based novelty detection, statistical inference based evaluation of learning, and probabilistic measures of prediction performance. These evaluation methods are illustrated as dynamic monitoring tools for two types of neural networks used in the NASA Intelligent Flight Control System (IFCS) as adaptive learners: the Dynamic Cell Structure (DCS) network and the Sigma-Pi network.

Computing is everywhere, and software is too, but what about the quality of this software? As software becomes increasingly pervasive in our society, what can we do to ensure its quality? One Software Quality Assurance mechanism, Software Testing, is often omitted from the development and implementation of computer systems, sometimes because of its perceived inconvenience and difficulty. A relatively simple way of testing software is to apply test cases (combinations of input representing a single use of the software) randomly, a method known as Random Testing. Among the advantages of Random Testing are its ease of use, the minimal overheads in test case generation, and the statistical support available. Some research has indicated that more widespread distributions of test cases throughout the input domain may be more effective at finding problems in the software. Adaptive Random Testing methods are Software Testing methods which are based on Random Testing, but which use additional mechanisms to ensure more even and widespread distributions of test cases over an input domain. This chapter gives an introduction to some of the major Adaptive Random Testing implementations.

The need for adaptability in software is growing, driven in part by the emergence of pervasive and autonomic computing. In many cases, it is desirable to enhance existing programs with adaptive behavior, enabling them to execute effectively in dynamic environments. In this chapter, we introduce an innovative software engineering methodology called transparent shaping that enables dynamic addition of adaptive behavior to existing software systems and applications. We describe an approach to implementing transparent shaping that combines four key software development techniques: aspect-oriented programming to realize separation of concerns at development time, behavioral reflection to support software reconfiguration at run time, component-based design to facilitate independent development and deployment of adaptive code, and adaptive middleware to encapsulate the adaptive functionality. After presenting the general methodology, we discuss two specific realizations of transparent shaping that we have developed and used to create adaptable systems and applications from existing ones.

Rule extraction as a specific technique can support the rigorous development and assurance needed before using adaptive systems in mission- and safety-critical applications. The ultimate goal and significant innovation of this research discussed in this chapter is to demonstrate that neural network rule extraction technology could be transferred into a practical software tool for neural network verification and validation and other purposes, such as monitoring the state of a neural network. Such a tool would accept as input a formal specification of the trained neural network and use neural network rule extraction algorithms to translate the neural network into an equivalent set of rules. These rule-based systems, which represent the neural network's knowledge, have a more visible and potentially human-readable decision logic that supports a robust set of verification techniques. Rule extraction technology in the form of a usable tool will dramatically increase the ability to verify and validate high assurance neural network systems.

Adaptation creates new challenges for software and system requirements engineers. If the system is expected to change its operation, how do we express and ensure the bounds of “desirable” changes, i.e., those that by some measure improve rather than deteriorate the performance? This paper introduces a technique for the analysis of boundedness problem for a general class of dynamical systems subjected to direct adaptive control using Lyapunov analysis, within an “adaptive augmentation” setting. An accurate formulation of the bounding set for the overall closed loop system is derived, under the assumption that the error driving the adaptive element is chosen as a static function (i.e., a matrix gain) of all the measurable outputs of the system. Based on this result, we present a simulation example relative to an adaptive flight control system designed for F/A-18 aircraft. Further, we discuss the impact that typical design parameters have on the shape and dimensions of the bounding set.

A software development process modeling framework is motivated and constructed under the formalism of State Modeling. The approach interconnects instances of general development-activity modeling components into a composite system that represents the software development process of the target organization. The composite system is then constrained according to the interdependencies in the work to be completed. Simulation results are presented, and implications for control-theoretic decision support (i.e., state variable control) are briefly discussed.

Systems based on service-oriented architecture are called service-based systems (SBS), and comprise of computing services offered by various organizations. Users of SBS often require these services to be composed into workflows to perform their high-level tasks. The users usually have certain expectations on the overall QoS (quality of service) of their workflows. Due to the highly dynamic environments of applications of SBS, in which temporary unavailability or quality degradation of services may occur frequently and unexpectedly, monitoring and controlling the execution of workflows adaptively in SBS are needed and should be done in distributed and proactive manner. In this chapter, important research issues and the current state-of-the-art will first be discussed. Then, a virtual machine-based architecture for the execution, monitoring and control of workflows in SBS, and a process calculus for modeling distributed monitoring and control modules are introduced. Using the virtual machine-based architecture and the process calculus, our approach to synthesizing software modules for proactive monitoring and control of workflow execution in SBS is presented.

Accelerated life test (ALT) methods are successfully applied in many industries to reduce the test period of highly dependable products. Software industry is not different, having the same demand to reduce the period of test for software products with very low failure rates. Since software is now part of many important processes of modern life, increasingly software-based products are expected to be highly dependable, which requires sophisticated techniques to test them within acceptable time frames.

The use of ALT assumes that the system under test has its life reduced when it is exposed to some stress load. Many software products show a systematic performability and dependability degradation under certain circumstances, mainly when varying workload and long execution time are present. These software system degradations have been investigated over the last decade and now they are well explained through the software aging theory. In this chapter, we discuss how to apply accelerated life test techniques to software systems that suffer from software aging. We show how the software aging theory enables the usage of ALT methods to estimate the time to failure of a software system. We show a real case study, related to a widely adopted web server software system, to illustrate how to systematically apply ALT in software experimental studies. In this experimental study, ALT offers a reduction of approximately seven times the time required to obtain the same amount of failure data in use condition (without acceleration).