Showing papers in "Isa Transactions in 1992"

Application of neural networks to water and wastewater treatment plant operation

Abstract: The application of artificial intelligence techniques to the operation of water and wastewater treatment plants in recent years is reviewed. The expert system approach is the most prevalent, but difficulties in acquiring and representing knowledge of the complex phenomena in these plants have led to the search for additional approaches. Fuzzy logic and statistical process control are used for formulating expert rules from plant historical operating data, but artificial neural networks, which can learn from examples, are believed to be a better solution for this task and for many additional problems encountered in the operation of the plants. Basic concepts of neural network organization and training are given as well as recent advances in learning speed improvement that have paved the way for easy application of this technique in large industrial plants. Current and future utilization of neural networks in areas of water and wastewater plant modelling, expert rule extraction, fault detection and diagnosis, plant and instrument monitoring, dynamic forecasting, and robust control are discussed. Examples are given from the application of neural networks to the operation of the Shafdan wastewater treatment plant in Israel. Some limitations of the neural network approach, together with ways of overcoming these limitations, are described. The overall conclusion is that we will soon see neural network techniques applied to achieve better plant operation.

Modeling human operators using neural networks.

Abstract: Often in the process industries, human operators rather than mathematically based advanced algorithms are used to achieve product control. Typically, if there are three shifts of operators per day, one shift achieves superior control. If the expertise of the best operator can be captrured easily and economically and made available to the other operators, significant economic benefits would accrue. This paper discusses a methodology that uses artificial neural nesworks for capturing the knowledge of process operators. For many operator tasks, only readily available information obtained from a process control computer is required. Once a converged network is available, a stripping technique can be employed to stimplify the net and to gain knowledge about what a good operator is doing compared to a poor one. It is felt that the proposed approach is superior to traditional expert system techniques employing knowledge estraction approaches.

Strategy of stochastic real-time control of wastewater treatment plants

Abstract: The paper describes fundamentals of stochastic real-time control (RTC) of wastewater treatment and disposal facilities operating under dynamic wet weather/dry weather conditions. Similar models are also applicable to the control of combined sewer overflows (CSO's). Objectives and goals, components, strategy, and operation of RTC systems are delineated. RTC systems can be incorporated into an expert system to assist the operator of wastewater treatment and disposal facilities, and/or into semior fully automated systems. The stochastic transfer functions used for prediction of the inputs and response of the system belong to the class of ARMA stochastic models. Neural network models are also applicable. Such models can be shown to be consistent with traditional deterministic transfer functions. A review of existing RTC systems concludes the paper.

Control of wastewater treatment systems

Abstract: Some basic ideas of process control applied to the activated sludge process are presented. The focus is rather on controller structures than on models. The reason for the control is due to disturbances and load chances. Due to the wide span of time scales of the process it is possible to realize many control actions decoupled from each other. Some of the controllers are quite simple and do not require any model, while others are related to quality parameters and are based on model descriptions. By combining simple controllers, more complex structures can be obtained, such as cascaded control and feedforward control. For some control tasks the simple controllers are inadequate. Processes with time-variable parameters present special problems. Since many variables and parameters are not directly measurable, it is valuable to apply estimation techniques in order to obtain indirectly calculated variables. The interdisciplinary character of the problems has to be understood in order to find relevant solutions to the complex control problems.

Information processing coupled with expert systems for water treatment plants

Abstract: The water utility industry is facing increased pressures to produce higher quality finished water at lower costs. The traditional feedforward-feedback control systems used in water treatment plants can be enhanced by using predictive models (with feedforward control) and diagnostic expert systems (with feedback control). The use of predictive models is advantageous as it is virtually impossible to ascertain specific chemical dose requirements from measurements of raw water quality parameters. These predictive models may be either regression relationships or neural networks. When feedback data indicate that the finished water quality is not acceptable and usual corrective actions based on feedback from sensors fails to correct the situation, diagnostic expert systems can be effectively used to aid operator decision making.

Case-based learning: A new paradigm for automated knowledge acquisition

Abstract: Case-Based Learning (CBL) is a new paradigm for solving problems by generalizing and transforming solutions of similar previously encountered ones. Cases serve as actual problem-solution instances in CBL. Here, cases differ from those in the Case-Based Reasoning (CBR) paradigms in that they are immutable. Expert n systems define a network of intercting domain-specific subsystems for bootstrapping CBL methods. The primitive subsystems are called next-generation expert systems. All subsystems constrain random generalization spaces. Generalizations can be augmented with a CBL method called Random Seeded Crystal Learning (RSCL). RSCL methods generate a population of similar hypotheses using a tranformational paradigm. The cases serve to delimit this space while simultaneously refining the process for generating the hypotheses. RSCL methods provide for the shifting of the knowledge engineer's focus from the design of rules to the design and bootstrapping of domain-specific languages for their capture. A prototype CBL system called LS has been realized at NOSC. LS efficiently induces novel rules that are open under deduction. Results indicate that more knowledge can be generated than is supplied. LS is implemented on a DAP-610 platform in approximately 10, 000 lines of FORTRAN-Plus. Its utility—especially as applied to intelligent manufacturing systems—is expected to mirror advances in parallel hardware.

The application neural networks in the process industry

Abstract: This paper will present an overview of the history and definition of neural networks. Information will be presented on current neural network commercial products and their applications. A case history of neural network technology used in a chemical process application will be discussed along with recommendations of where neural networks may most effectively be used in the process industries.

The role of integrated AI technologies in product formulation

Abstract: This paper describes a technique for integrating expert systems, optimizing techniques, and proprietary neural networks to vary the formulation and processing of products. This is done to achieve new product properties or satisfy new cost, processing, or property constraints. The formulation of products such as rubber, plastics, alloys, and pharmaceuticals is addressed. The paper evidences the importance of automation in creative design.

Expert systems in process control

Abstract: This paper discusses the history of expert systems, their migration to the process control room, and their application in process industries. Two case studies that discuss the justification, development, and implementation of expert system applications are presented. The authors' views on the future of expert systems in the process control environment are introduced in the Summary/Conclusions section of the report.

Knowledge-based expert system applications in waste treatment operation and control

Abstract: Experience has shown that operation of wastewater treatment processes can be difficult. In some plants, poor performance, high costs, and damage to the environment have resulted from operational problems. Progress in control engineering, computer technology, and process sensors has enabled development of computer-based systems for automatic control and operations support in wastewater engineering. In the area of automatic control, important advances have been made; however, because of the unique character of wastewater treatment, techniques other than a straightforward application of control theory are needed to achieve effective control. Preparing computer-based tools to aid in operation is a modelling task, with the primary objective being to represent knowledge about a process and its operation. The traditional approach is to use mathematical methods for this purpose. More recently, tools and techniques from applied artificial intelligence have been used to prepare systems that encapsulate knowledge of diagnostic and control decision tasks. These knowledge-based (expert) systems (KBES) show promise as aids in support of such tasks. Several KBES development projects have been reported with varying degrees of success. There is a high level of interest in KBES, but few full-scale systems have been reported. Most of the documented problems concern initial elicitation/capture of knowledge and final verification/validation of the KBES. Recently, improved KBES have been reported including object-oriented systems, systems using deep knowledge representations for qualitative reasoning, and integrated systems that make use of numeric and symbolic processing.

Instrumentation control and automation in the control of biological effluent treatment.

Abstract: The role of instrumentation, control, and automation (ICA) in the management, operation, and control of domestic and industrial wastewater treatment works is examined following an historical account of developments in the UK over the past two decades. Included also are aspects of process optimization and process selection, taking into account size and types of works and populations served. The particular ICA requirements of individual unit processes are examined and the effects of unattended operation and remote supervision are considered. Particular attention is paid to measurement technology, the shortcomings of available instruments and systems, and recent developments of relevance, with particular emphasis being placed on microelectronic and fibre- and electro-optic-based sensors and on the role of advanced commenting methods including expert systems. Progress with biologically based sensors is also discussed.

Emerging chemical sensor technologies for use in on-line applications

Abstract: Advances in instrumentation and process control capabilities are being driven by the increased sensitivity of analytical chemistry methods, the commoditization and ruggedization of instruments and associated electronics, and changing consumer concerns and expectations, as well as government regulations. The emergence of biotechnology methods for the detection of contaminants, and the miniaturization and ruggedization of more standard analytical instruments based on mass spectrometry and infrared spectroscopy will be discussed as related to on-line process monitoring.

Next generation manufacturing task planner for robotic arc welding

Abstract: The trend toward manufacturing in small lot sizes continues, imposing ever new demands on the responsiveness of companies to consumer demands. To be cost competitive, this requires the designer to perform part of the manufacturing planning early in the design phase. Towards this end a number of decision support tools have been developed, providing designers with vital information regarding the manufacturability of a part or a product. Part of any manufacturability study includes the simulation of the processes involved. It is important that the designer can investigate different manufacturing scenarios and successfully choose one within the constraints of his organization or the ones he can easily access. In this paper we describe the requirements of an environment in which a part can be designed incorporating both the geometric and nongeometric information required for a complete product/manufacturing process description. This paper also describes the nonprocedural Programmable Automated Welding System (PAWS) package. This system has three major components, namely, a non-manifold solid modeler for part, equipment, and environment definition, an SQL (standard query language)-compatible database for process definition, and a robotic motion planning module capable of motion optimization, interference detection and graphic simulation. Although this system focuses on the welding process, the key components of the system are equally relevant to other manufacturing strategies such as grinding, spraying, deburring, and cutting.

Generalized multiple-regression techniques with interaction and nonlinearity for system identification in biological treatment processes.

Abstract: A class of multiple regression models, called "generalized multiple-regression" (GMR) is proposed. GMR has the advantages of being easy and rapid to fit, and uses standard multilinear regression software. It has an advantage over ARIMA models in modeling nonlinearity and linear and nonlinear interactions among variables. Its main disadvantage is that, if there are many independent variables, the reduction of degrees of freedom may be important. It is less parsimonious than other models, but availability of increased computational power makes this not a serious disadvantage. The GMR models are compared to autoregressive transfer function models and feedforward back propagation neural network models. In the case of modeling effluent volatile suspended solids, GMR models were superior to both linear autoregressive models and neural network models. The neural network models did, however, outperform the linear models. In the case of modeling sludge volume index, both GMR and the neural network model were unable to improve upon ARIMA models. It was concluded that ARIMA models may, in some cases, produce the most parsimonious model, but in other cases they may miss important process behaviors. The GMR models showed robust capability to describe complex data.

Expert systems applications in quality management: An integrative approach

Abstract: Quality management is a vital and growing concern of businesses in the global marketplace. Expert system technology is beginning to play an important role in this area and will become much more common in the future. This paper summarizes several current expert systems used in quality management. A framework for choosing appropriate application candidates and a guide to integrating expert system technology into these applications is presented. Following the framework will significantly improve the chances for a successful expert system installation.

Engineering design support framework by case-based reasoning

Abstract: In engineering design, the customer's order-bandling process normally requires the designer's previous experiences. The designer may be required to modify basic product units as well as install new functions. We propose a case-based design support framework to support this process. Our framework is implemented as a system SUPPORT, which interactively supports the design process involving the specification analysis, the functional design, and the parts selection and layout. The designer can analogically reuse past designs and/or their pieces from various viewpoints. After a new design is completed, the functional design is organized into an existing function abstraction hierarchy and can be used as a framework for a new design in the future.

Programmable logic controllers find a home in wastewater treatment

Abstract: Efforts to computerize wastewater treatment plants have had mixed results. A wastewater treatment plant presents significant obstacles to computer automation. One of the most challenging problems is coping with the aggressive environment of the plant. Industrially hardened programmable logic controllers (PLCs) have found a home in the wastewater treatment field. These small, reliable, and rugged computers are gaining acceptance in this conservative industry. Standard process control and data acquisition software packages are available from most PLC manufacturers for use on a supervisory personal computer workstation networked to their PLCs. In addition, similar third-party generic software packages, which support the PLCs of multiple manufacturers, can be procured. These packages provide functions such as interactive graphics, database, trending, historic storage, and report generation. Networks containing personal computers and PLCs offer an attractive alternative to the conventional large distributed computer system (DCS). A compomplex PLC network can equal a DCS in capability, while a simple PLC network can offer fast, reliable control of a single process. In each case, PLCs offer the advantage of a rugged input/output (I/O) system designed for the factory floor. Process control applications with fail-safe provisions are suitable to PLCs and can be demonstrated throughout the plant. Although they are industrially hardened, caution must be exercised in the application of PLCs. The corrosive nature of the atmosphere at a wastewater treatment plant still presents challenges. However, properly installed, PLCs maintain their reputation for reliability.

New technologies in infrared hydrocarbon detection

Abstract: For the last three decades electrocatalytic sensor technology has served as the “standard in the industry” for the detection of potentially combustible levels of hydrocarbon gases. While other technologies have existed for years that have offered alternatives to the electrocatalytic hydrocarbon sensor, electrocatalytic technology has held fast as the technology of choice as these alternative technologies either proved unreliable or prohibitively expensive to produce as a field-mounted sensing device. Recent advances in the field of infrared technology have resulted in the development of practical infrared hydrocarbon detectors that allow for significantly improved reliability, area coverage, and drastically reduced maintenance requirements of refineries, petrochemical plants, and gas processing plants, among others. The history, current situation, and possible future of infrared gas detection are reviewed.

Real-time advisory control applications in the petrochemical industry

Abstract: The petrochemical industry has benefitted from expert system technology in many ways. It provides a tool for allowing companies to capture operational knowledge, distribute it over time and applications, and provide advice for predictive and corrective action in the plant. However, plants have found that to meet the demanding requirements of today's complex CIM environments they need more than what is available with current off-line and real-time expert systems. Major needs include full integration with existing and future control systems, real-time response, robust performance in a plant environment, and the flexibility to present advice on existing control system and information system screens throughout the plant. This paper discusses these industry needs and the limitations of current real-time expert systems and presents successful applications of RTAC, a real-time advisory control system. These petrochemical applications illustrate how real-time advisory control software helps customers meet their needs for both proactive and diagnostic on-line advice in an integrated CIM environment.

Expert systems come to batch process automation

Abstract: Batch process control systems tend to be highly sequential and structured. There is a premium on predictability and repeatability since these lead to a desired uniformity between batches of a product. Conversely, expert systems tend to be unstructured and somewhat unpredictable because of the variety of ways in which a new piece of information, or the need to answer a question, can lead to the use of rules or to other actions. At first these fundamental structural differences might be seen as an impediment to the widespread use of expert system technology in a batch automation environment. However, the expert system approach to problem solving is shown here to complement traditional batch-automation techniques in several applications. Additionally, the integration of procedural and expert system capabilities into one software application environment can help make the advantages of each equally accessible to the control engineer.

Information management for wastewater treatment

Abstract: The data collected on a wastewater treatment plant should not only describe its current condition, it should also be used to continually improve the operation of the system. This is possible only if the right information is displayed in ways that allow for its absorption and analyzed in terms that are useful for recognizing and anticipating problems. Experimentation is important in elucidating the behavior of a system in response to perturbations in inputs and operating conditions. This dynamic behavior is seldom revealed by analyzing happenstance data. This paper describes several statistical methods that are useful for information management. These range in difficulty from simple plots (one of the most important tools) to time series modeling for detecting trends and changes in level, and for constructing control charts that satisfy the basic statistical requirements. A new method called discriminant upset analysis is proposed as a possible tool for anticipating upsets. How these methods might be incorporated into an expert system is outlined. It is more and more often necessary to monitor and control chemicals that exist in trace amounts, sometimes at concentrations that are too low to be measured by existing analytical methods. As a result, many data sets are censored at the analytical limit of detection of the substance of interest. This presents special data management and analysis problems, and some suggestions are made regarding their solution.

On automating goal-to-task translation in a futuristic robotic factory☆

Abstract: A major technological impediment to increasing productivity in the manufacturing sector is the relative absence of automation at most stages of the product design process. One such stage involves the translation of non procedural goal files—describing the desired result of some aspect of the manufacturing process — into procedural task files. The task files detail the exact sequence of required steps as typified by an assembly operation. These steps can be effected by cooperating industrial robots. Therefore, it is important that the task files provide for coordination among the robots and minimize costly downtime. This paper advances a technique for automating the labor-intensive translation of goal-to-task files. The process entails the development and modification of an expert translator. Modification is part of a user-interactive on-line feedback loop or learning system. The system has been implemented at NOSC in C and is supported by the Sun 3/50 platform. It is currently undergoing modification for application at CMU's futuristic robotic factory.

New sensor developments

Abstract: The new sensors are the result of new materials, techniques, and calculation and decision capabilities. They are more accurate, reliable, rugged, noncontacting and digital. A microprocessor is integrated with nearly every sensor system, giving both local and milewide access to all information. Many of the new sensors are basic in function — smell, taste, tactile, machine vision, and sound. For example, new odor meters indicate temperature rise as the odorous gas oxidizes or due to an electrode conductivity changes; guidelines are established for quality standards or acceptable odor levels. Or the outputs from several sensors of variables associated with the flavor of liquid are combined in a trial and error relationship until a satisfactory correlation is achieved by artificial intelligence. Thin-film sensors for tactile measurements are based upon electrical conductivity changes with pressure; microprocessor scanning of electrodes results in a quick display of sensitive field-effect transistor pH meter for a drop of sample; chemical environmental sensors; and a micro-gas chromatograph. Fast Fourier Transforms (FFT) applied to a Doppler flowmeter enhance the speed and sensitivity by a factor of ten. Self-calibrating sensors with corrections for either variables include a new radiometer with laser-corrected emissivity and a self-calibrating fiber optic temperature sensor insensitive to emissivity. Infrared photoacoustic sensing is the newest generation of advanced technology sensors for gas analysis, and it results in a most stable zero and span. Other noteworthy new products include hybrid photodetectors, solid-state toxic gas detectors, ultrasonic and microwave radar level and distance gages, microminiature color camera, color scanner, surface quality meter, and multi-wave length spectral analyzers.

Technology-based education and training during the 1990s

Abstract: Training is a multibillion dollar cost to industry. It is estimated that $30B is required just to bring the entry level skills of new employees up to high school diploma equivalency. Traditional training is a continuous labor-intensive process. Basic computer-based training (CBT) has proved to be lacking and in many cases not more pedagogically effective than an electronic book or reference manual. Education in this country continues to fall behind what is required to sustain high-technologybased industries. Educators are overwhelmed in trying to maintain knowledge of the state of the art in key technologies. Without a doubt, new technologies will of necessity become an integral part of all education/training programs. Human resources to supply populations of trainers will generally prove too costly or scarce. Further globalization of the economy will further separate the human expert from the trainee. We will be afforded a wealth of enabling technologies for education and training during the 1990s. Not only will we be challenged to keep up with them, but harnessing them to our best use will be a critical issue. This article will explore some of these technologies and the surrounding issues.

A knowledge-based approach to programming welding robots

Abstract: The article describes ROBOEDIT, a knowledge-based programming assistant that has been developed to facilitate the modification and the fine tuning of robot programs on the production floor. In particular, ROBOEDIT is defined for use with the Advanced Robotics Cyro 5+2 are welding robot. The system is written in LISP and runs on a microcomputer under MS-DOS. It employs several knowledge-engineering principles to improve production floor editing efficiencies and to simultaneously reduce robot programming skill requirements. The latter goal is accomplished largely via a user interface that features an English-like format. Parsing techniques based upon context-free grammars are used to translate robot programs between the user interface and the robot language syntax. An indexed database design is employed to build representations of the information contained in the programs and to encode important relationships. Finally, production system formalisms and robot-specific knowledge about kinematics and programming rules are combined to create an integrated environments that assists the user during editing sessions and check edited programs for correctness of syntax. A sample session with ROBOEDIT is described and discussed.

The personalization of manufacturing: Redefining the human interface

Abstract: How do manufacturers compete in a world where the sheer volume of global markets, as well as the growing number of “custom products,” places the organization is a real dilemma—how to concentrate on providing fast turnaround while preserving high quality standards in job-shop, batch, and volume production situations? We will focus on the newest era of process innovation, the “personalization of manufacturing” and how graphical computer systems are being used to redefine the computer interface to manufacturing processes and personnel. These inexpensive, yet powerful distributed systems are now able to provide manufacturing operations with a significant advantage in terms of the operator interface and still offer powerful HO capabilities. In fact, the progress made by these systems promises to soon bring about the day when factories are filled with computers that adapt to their users, as opposed to computers that require their users adapt to the technology.

An object-oriented operator's interface for real-time process control expert systems

Abstract: Critical to the success of real-world process control expert systems is acceptance by the operator. An operator's interface that is native to the existing control system and that utilizes existing protocol has a clear edge over an interface presented on a foreign device. This paper describes the design, functionality, and related ergonometric issues of an object-oriented operator's interface deployed for real-time process control expert systems. Native to the existing control system, this interface-and the expert system behind it—represents an extension of the existing control system rather than a departure from it. Because this system utilizes existing process graphics, touch-screen interaction, and other “standard” control system features, it results in u less threatening, more acceptable introduction of expert system technology into the control room.

Kiss your PLC

Abstract: The sage adage, “ Keep it small and simple ,” provides a memorable acronym (KISS) and good advice for any venture involving complex computers. Yesteryear efforts to incorporate this technology into the traditionally conservative realm of water and wastewater engineering all too often overlooked this KISS approach, however, falling headlong into an electronic abyss that still haunts those plants that strayed too far too fast. Hence, the field has recently begun to shift downward in scale and sophistication, towards a hardware grouping largely dominated by a low-end product known as a programmable logic controller (PLC). Indeed, these PLC's systems appear to offer a prime opportunity for small-to-medium scale water and wastewater treatment facilities to explore a wide range of monitoring and control regimes with minimal financial risk…fully within the KISS context! This paper will, therefore, explore the conceptual development of three small-scale PLC applications that largely emphasize a process oversight or “caretaker” role, but which also have clear potential to incorporate additional missions and responsibilities relative to their affiliated control requirements.