Power Supply Circuit Development Estimating Aid
An Expert System Application

Jerrold Foutz
Autonetics Electronics Systems
Rockwell International Corporation
Anaheim, California 92803-4192

Abstract

An expert-system-based circuit development estimating aid generates a cost estimate based on the work-plan method. Outputs include hours by labor category, computer and material dollars, and critical path scheduling information. A panel of engineers with over 200 years of power supply design experience provide the expert knowledge used.

Introduction

The purpose of this work is to develop a computer-based estimating aid that will allow a power supply circuit designer to rapidly generate a consistent and defendable circuit development cost estimate. Perhaps more important are its additional uses. The aid provides insights into how the cost of power supply circuit development can be reduced, and serves as a training and control tool for a disciplined approach to power supply circuit design.

The approach is to use a computer-based expert-system shell to design an expert system that uses expert knowledge embedded in If Then rules. The expert system then assembles work plan subelements into a power supply circuit development estimate.

The intended user of the expert system is the power supply designer doing conceptual design in support of proposal activity. In response to inputs provided by the designer, the expert system generates a cost estimate that includes labor hours, material and computer-time dollars, and a critical path schedule.

The expert system is structured to provide insights into opportunities for cost reductions. This can be done by modifying the work statement to eliminate subelements, or by increasing the efficiency for doing each subelement by better use of tools or training. The expert system has to be modified to reflect these changes, but easy modification is one of the strengths of the expert-system approach. In today's competitive environment, this is considered an important feature of the program.

The initial calibration of the knowledge system was by a panel of engineers with over 200 years of power supply circuit design experience. The expert system also provides for the use of experience curves and historical data to refine the estimating accuracy.

Although specifically designed for power supply circuits, the expert system can be easily expanded to encompass any analog circuit. A similar expert system could be developed for digital circuits.

Expert Systems And Shell Selection

Expert Systems

Expert systems are an early and successful application of artificial intelligence. For an expansion of the following summary, the reader is referred to one of the many books on artificial intelligence or expert systems such as the introductory book by Harmon and King (1). The basic components of an expert system are a knowledge base, an inference engine, and a user interface. An expert-system shell is an expert system with the knowledge base removed, leaving the inference engine and a user interface.

Knowledge Base.

Knowledge can be represented in the knowledge base in several ways. One of the most common methods is a rule-based system where the knowledge is represented in IF/THEN rules. For example:

  IF:       circuit type is switching regulator  
  THEN:     circuit classification 
            is SWITCHING MODE CIRCUIT 
            and REGULATOR CIRCUIT  
            and FEEDBACK CIRCUIT 
            and ALL CIRCUITS

If the conditions of the premise in the IF part of the rule are true, then the conclusion in the THEN part of the rule is true, or the action called for in the THEN part of the rule is executed.

Inference Engine.

The inference engine is a method of operating on the knowledge contained in the IF/THEN rules. Backward-chaining inference engines are the most common. In a backward-chaining inference engine, a goal or choice is defined and placed in the THEN part of a rule. The inference engine takes the THEN part of the rule as a goal, or hypotheses to be proved, and checks to see if the conditions of the premise in the IF part of the rule are true. The inference engine chains to other rules as necessary to reach a conclusion about an IF premise, eventually asking the user if the information cannot be found from rules. For example, in the preceding rule, in trying to determine the circuit classifications, the expert system would ask the user the type of circuit if this was not already known to the system.

User Interface.

There are usually two interfaces to an expert system. One interface is for the developer of the system and one is for the consumer. Both interfaces usually include an explanation facility that explains the facts and reasoning used by the expert systems in reaching its conclusions. The developer's interface creates a sophisticated programming environment that enable rapid prototyping and development of an expert system. The developer's interface is also used for maintaining the expert system. The consumer's interface provides a user-friendly interface for the final user.

Advantage of Expert-system Shells

The estimating aid described in this paper could have been written using conventional programming techniques. An expert-system shell was selected because it offers several advantages.

The sophisticated user interface allows the engineer to concentrate on the knowledge content of the program instead of input, output, and comments. With the entry of the first top-level rule, the program is up and running, asking for information from the user if the information can not be derived from the rules. The program continues to run as rules are added. The explanation facility allows for easy debugging if the answers are incorrect. The sequence of the rules is usually unimportant, eliminating the sequential discipline needed in conventional programming. The approach for solving the problem does not necessarily have to be known at the beginning. By experimenting with rules and observing the results, the program often reveals a useful method for solving the problem.

For these reasons, the author views an expert-system shell as a productivity software tool similar to a word processor, database, or spread sheet; and believes it only a matter of time before expert-system shells are used as routinely as these other software productivity tools.

Selection of the Expert-System Shell

Over 60 potential applications of expert systems in support of the power supply circuit designer have been identified (2). The requirements of several of these expert systems were used to establish the requirements for the selection of an expert-system shell. Many of these applications have limited domains and useful expert systems could be developed with one to four person-weeks of labor.

User.

The power supply designers throughout Rockwell International Corporation (Rockwell) are potential users of the expert systems being considered. Any individual expert system could have from one to over 200 users. Rockwell's customers are also a potential user of some of the expert systems.

Hardware.

All of the potential users have access to IBM PC or compatible computers with a minimum of 512K memory, two disk drives or a disk drive and hard disk, and CGA graphics capability. This was selected as the target machine. The expert-system software had to run at reasonable speed on the target machine without the need for any software except that delivered with the expert system.

Run Time License.

The cost of distributing the expert system is a critical selection criteria and is one that is often overlooked. The selected expert system must have a low cost, hassle free method of distributing complete expert systems. A one-time fee that allows for distributing unlimited copies of an unlimited number of expert systems would be preferred.

Initial Cost.

Since the expert-system shell is considered a productivity tool similar to word processor and spread sheets, the initial cost should be compatible with this type of productivity tool.

Features.

A rule-based system is desired, since this is a natural method for engineers to state much of their knowledge. A backward-chaining inference engine is also desire, although capability to do forward chaining is an additional benefit. The shell must provide easy interfaces to outside programs, and be capable of displaying large text files. Graphics capability would be a welcome additional feature.

User Friendly.

The goal is for the domain expert (the power supply designer) to be able to incorporate knowledge into the expert system without the need for a knowledge engineer (an expert in representing knowledge and processing it be artificial intelligence techniques). Therefore, the shell should not require any deep knowledge of programming or the teaming with a knowledge engineer. The shell should require less than a week of training and experimenting for a user to become effective in its use.

EXSYS Shell.

At the time the requirements were defined in early 1983, no shell was found that met the requirements. In 1985, EXSYS was identified as a shell having most of the desired features (3). In early 1986, the shell was purchased along with a runtime license. Although other expert systems are now available that would meet the requirements, EXSYS has worked well in a variety of power supply applications and is the shell used for the power supply circuit development estimating aid.

The User And User Environment

The intended user is the power supply designer. In the proposal environment in which the aid will be used, the designer's first task is to develop a conceptual schematic of the circuit with a generic parts list that will meet the customers requirements. This information is used in a hardware description that allows manufacturing operations to estimate production costs, and allows other organizational functions to estimate their costs.

The designer next prepares a cost estimate for developing the circuit and other tasks in the statement of work. The workplan method, where each task is individually planned in detail, is a preferred method of estimating for several reasons, including the fact that detailed planning is already accomplished at contract award. However, the work-plan method is the most time consuming approach and there is rarely time or funding for developing a detailed work-plan as part of the proposal activity.

The power supply circuit development estimating aid circumvents the time and cost constraints of doing a work-plan estimate, An estimate based on detailed work plans can be developed for each circuit in a few minutes time.

After preparation of the hardware description, the designer knows the circuit type, part types (including magnetics), part count, operating environment, and contractual requirements for analysis, test, and documentation. This information is used by the expert system to generate the circuit development cost estimate.

The output includes total labor hours, broken down into the categories of engineer, technician, and magnetics design; material dollars; computer-time dollars; and a critical path schedule. These outputs are compatible with other programs that collect the total costs for a proposal. In addition to the main outputs, backup information is provided to support the cost estimate.

SAMPLE CONSULTATION

An example dialog illustrates how the estimating aid interacts with the user. Fictitious cost data is used, since actual cost data is considered proprietary information. The dialog starts by displaying initial information.

POWER SUPPLY CIRCUIT DEVELOPMENT
ESTIMATING AID

This estimating aid generates a circuit development cost estimate for power supply circuits by assembling work-plan subelements using built in knowledge of what subelements are required for specific specifications and circuit types. The assumption is made that the designer has done one or more similar designs. If this assumption is not correct, the estimate must be modified by a learning curve, either manually or by another expert system. The aid is intended for use by the circuit designer immediately after providing inputs to the hardware description. At this point in the estimating cycle, the design specifications, selected circuit topology, and parts list are available. This information is used to generate the cost estimate. The aid also generates a critical path schedule for the circuit development.

Please input name of the circuit
:5V REGULATOR

type of circuit is
1 series regulator
2 zener regulator
3 switching regulator
4 tr filter set
5 input filter
6 other
: 1

Please input number of transistors
: 6

Please input number of diodes
: 3

Please input number of integrated circuits
: 0

Please input number of magnetics
: 0

Please input number of capacitors
: 8

Please input number of resistors
: 16

Please input number of circuit inputs
: 2

Please input number of circuit outputs
: 1

Please input number of nodes
: 17

Please input number of feedback loops requiring stabilization
: 1

Please input number of operating modes
: 1

Please input number of devices requiring special heat sinking
: 1

Please input upper temperature extreme in degrees C
: 125

Please input lower temperature extreme in degrees C
: -55

Please input parts list cost in dollars
: 50

critical parts are
1 in stock
2 to be ordered
WHY
IF:
critical parts are in stock

THEN:
{DM] IS GIVEN THE VALUE 3*[DPL]
and top level estimate completed are material dollars

ELSE:
{DM IS GIVEN THE VALUE 5*[DPL]
and top level estimate completed are material dollars

NOTE:

If the critical parts are in stock, three times the parts list cost {DPL} is allocated for material dollars {DM}. The factor of three accounts for small quantity purchases, taking advantage of price breaks, misorders, and replacement parts. If the critical parts are not in stock, additional dollars are allocated to bracket long-lead-time parts to reduce the risk of not having the needed parts for breadboarding.
:1

Rather than immediately answering the last question, the user asked the question WHY. The system responded by displaying the rule whose conclusion was being sought, including a note explaining the rule. The user could then request additional reference information on the rule and it would be provided if available in the system. A WHY at any point in the program will display the rule chain whose truth is being tested.

The program then displays the results:

1. estimate complete
2. name of circuit = 5V REGULATOR
3. total hours = 513
4. engineer hours total = 431
5. Technician hours total = 82
6. magnetics design hours = 0
7. material dollars = 150
8. computer-time dollars = 231
9. requirements ltr complete week = 2
10. paper design complete week = 3
11. design complete week = 11

By typing in the number preceding the output, the program will respond with the rules used in deriving the answer.

The program can also be asked what it knows by asking for known data. In this example, 115 items would be displayed. Included is the information that the user entered and all the information concluded or calculated through application of the rules. Eleven of the 115 items that the system knows are displayed above. A report generator allows displaying any or all of the items in any order and desired form. The output can also go to a file that can be used as the input to another computer program or expert system.

DESCRIPTION OF THE PROGRAM

Overview

The Structure of the program is shown in Figure l.

GOAL ESTIMATE COMPLETE
OUTPUT TOTAL HOURS
ENGINEERING HOURS
TECHNICIAN HOURS
MAGNETICS HOURS
MATERIAL DOLLARS
AUTOCOMP DOLLARS
BACKUP DATA
SCHEDULE LEVEL PRELIMINARY ACTIVITY LEVEL
PAPER DESIGN
BREAD BOARD
MEASUREMENT
ANALYSIS
RECONCILE
DOCUMENT
SUPPORT
SELECTION LEVEL (SPECIFIC CIRCUIT) PRELIMINARY ACTIVITY
SERIES REGULATOR
ZENER REGULATOR
SWITCHING REGULATOR
TR SET
INPUT FILTER
OTHER
SELECTION LEVEL (GENERIC CIRCUIT) ALL CIRCUITS
REGULATOR CIRCUIT
FEEDBACK CIRCUIT
SWITCHING MODE CIRCUIT
FILTER CIRCUIT
SUBELEMENTS
ALGORITHMS
120+
120++

Figure 1: Program Structure

At the lowest level are subelements of the work plan that describe a detailed task that is done in developing a power supply circuit. For example, a task might be to order long-lead-time parts for the breadboard. At this stage in its development, the expert system works with about 120 such tasks.

Attached to each subelement is one or more calculation algorithms that calculate the hours and dollars associated with the task in the subelement. For example, to measure each node voltage on a breadboard, and record each measurement in a laboratory notebook, the algorithm may allot 12 minutes for set-up-time and 3 minutes per node for measurement and recording. To use this algorithm, the expert system must know the number of nodes in the circuit and will ask the user for this information.

Not all subelements apply to all circuits, and some means of selecting the applicable subelements is required. A generic grouping of subelements is made first. Some subelements apply to all circuits, and these are grouped together under the category ALL CIRCUITS. The category REGULATOR CIRCUIT would contain subelements such as line and load regulation measurements that apply only to regulator circuits. The category FEEDBACK CIRCUIT contains subelements such as analytical and measured open-loop gain and phase plots that apply only to feedback circuits.

The next selection level is controlled by the specific type of circuit being estimated and the user is asked to specifically identify the circuit type. This information is used to select the generic grouping of subelements that apply. For example, if the user identifies the circuit being estimated as a series regulator, a rule concludes that the subelements in the generic groups of ALL CIRCUITS, REGULATOR CIRCUIT, and FEEDBACK CIRCUIT apply. If necessary, a rule at this level can bypass the generic grouping and go directly to one or more subelements.

At this point, the subelements could be added together and the estimate generated. However, no scheduling information would be provided. Instead, the subelements are collected into categories related to a sequential development of the circuit. These categories are processed to generate the critical path milestones. This is useful in fitting the circuit development schedule into the proposal master schedule and in determining the resources needed to support the master schedule.

At this time, the outputs sought by the estimating aid are known, and the top level goal of "estimate complete" is reached. The report generator then outputs the information, along with backup data, in the desired format.

The expert knowledge in the program is embedded at several places. Expert knowledge is in the subelement descriptions that define each circuit development task, the algorithms that calculate hours and dollars for each task, and in the IF/THEN rules. These rules collect the subelements into groups applicable to generic and specific circuits and add them in a manner that generates a critical path schedule as well as total hours and dollar.

Work-plan Subelements

The work-plan details are contained in subelement descriptions as shown in the example in Figure 2.

SUBELEMENT DESCRIPTION

DESIGN PHASE: paper design
SUBELEMENT NAME: calculate component values
VARIABLE NAME: [HECCCV]
DATE: 10-07-87 REV A DATE: 11-04-87
APPLICABILITY: all circuits

ACTIVITY: Calculate value, wattage, voltage rating, etc., of all components. Prepare form 837-C "Request for Item Specification" or equivalent for any parts not on the standard parts list.

OUTPUT: Schematic and parts list with values and Form 837-C's.

ALGORITHM:

 
     [HECCCV] =        engineering hours for paper design
     0.0+              set-up time
     1.0*              time per part
     [PARTS COUNT]     circuit parts count

Figure 2. Subelement Description

The subelement description provides a standard frame with slots for incorporating information about a subelement.

The first slots contain the schedule grouping to which the subelement belongs; a short name for the subelement; the variable names used by the expert system for the subelement; and the origin date, revision letter, and revision date. The applicability slot provides the generic group of the subelement.

The activity and output slots describe what activities are included in the subelement and what are the outputs of the activity. These portions provide a detailed statement of work for the engineer and a check list for management to insure that the work is completed.

The algorithm slot provides a slot for the equation used to calculate the subelement cost and a description of each term in the equation.

A slot for cases contains one or more cases used to construct and evaluate the expert system. This slot also contains the results of a calculation of the subelement cost for the case. This feature provides a reasonability check as the expert system develops.

A final slot (not shown) provides a place for comments or discussion.

The subelement frames are kept in files with the same name as the variable. Presently, the files can be examined outside the expert system. In a later enhancement, these frames may be made part of the explanation portion of the expert system.

Calculation Algorithm

The most common calculation algorithm is shown in Figure 3.

The first part of the cost equation is a constant, called set-up time in the subelements, which accounts for set-up time, tear-down time, and other constant costs associated with the activity. For example, obtaining test equipment from the test equipment library and returning the equipment is a constant independent of the number of measurement made.

Example Calculation

Cost = Constant + Slope * Count

Figure 3. Example Calculation Algorithm

The next part of the equation is a slope multiplied by a variable. The variable is usually some count, such as circuit part count, number of feedback loops, number of inputs and outputs, etc. For example, the cost to calculate and record the wattage in each resistor may be 6 minutes per resistor multiplied by the number of resistors in the circuit.

More complex algorithms are used as required. For example, the number of one-at-a-time regulation plots for a regulator is the product of the number of inputs and outputs plus the number of outputs.

Example Rules

A backward-chaining expert system starts with one or more goals or choices and tries to validate each goal by checking its premises. The estimating aid has one goal to validate, "estimate complete", and this drives the backward-chaining inference engine to generate the estimate. This goal is stated in the rule:

IF:
     top level estimate completed is hours
     and top level estimate completed is material dollars
     and top level estimate completed is computer-time dollars
     and top level estimate completed is schedule

THEN:

     estimate is complete -- Probability = 10/10

The expert system than tries to validate the first condition in the premise of IF part of the rule and looks for a rule with the conclusion in the THEN part of the rule that states "top level estimate completed is hours". The system will find the rule:

IF:  
     second level estimate complete is preliminary activity
     and second level estimate complete is paper design
     and second level estimate complete is breadboarding
     and second level estimate complete is measurements
     and second level estimate complete is analysis
     and second level estimate complete is reconcile
     and second level estimate complete is document 
     and second level estimate complete is support

THEN:

     [HE] IS GIVEN THE VALUE [HEP]+ ...
     and [HT] IS GIVEN THE VALUE [HTB]+...
     and [HM] IS GIVEN THE VALUE ...
     and [H] IS GIVEN THE VALUE [HE]+[HT]+[HM]
     and top level estimate complete is hours

The expert system reaches back to find all of the applicable subelements that apply to hours, calculates the hours based upon the algorithms associated with the subelement and brings them forward where they can be summed in this rule as [HE] (Hours Engineer), [HT] (Hours Technician), and [HM] (Hours Magnetic), and finally summed as the total number of hours for developing the circuit [H]. The last statement makes the first condition of the goal-rule premise (top level estimate completed is hours) true and starts the system trying to prove the next condition (top level estimate completed is material dollars) of the premise.

In reaching back for subelements, the system collects the information in special subgroups (preliminary activity, paper design, etc.) that are associated with the time flow of the design process. This manner of collection allows a critical path schedule to be generated.

The system continues to backward chain until lower level premises are satisfied and all subelements applicable to preliminary activity have been identified. The system then collects the subelements for paper design, etc. This collection method allows other rules to construct a critical path schedule.

Another class of rules selects the appropriate subelements. The following rule identifies the generic grouping of subelements that applies to a series regulator.

IF:        circuit type is series regulator

THEN:      circuit classification is ALL CIRCUITS 
           and FEEDBACK CIRCUIT 
           and REGULATOR CIRCUIT

Other rules collect the subelements that belong to 
generic classifications such as all circuits.

IF:        circuit classification is ALL CIRCUIT

THEN:      ...(take action to invoke all algorithms that apply to ALL CIRCUITS)

The total expert system consists of rules similar to the preceding. The order of the rules is unimportant.

SOURCES OF KNOWLEDGE

Single Expert Problem

The normal process in developing expert systems builds the expert system using the knowledge of a single expert. While this is valid for an estimating aid, building the system this way can be challenged by bringing in another expert and asserting that "my expert is better than your expert".

Former Solution

At Rockwell, this problem was addressed in the mid 1980's by one power supply group. Several experienced circuit designers judged the cost of developing power supple circuits of various levels of complexity. Five sample circuits of increasing complexity were used as standards. The average of the estimates for each circuit was taken as a standard estimate for that complexity of circuit.

Cost estimates were than made by matching the complexity of the circuit to one of the standard estimates and using the standard cost for that complexity. The method was calibrated by feedback from the financial records on programs using the standards. This method was used for about a decade and then fell into disuse.

Interim Solution

It was planned to reactivate and recalibrate the old standard and write an expert system to determine which of the five circuit complexities applied to a circuit. It took less than a day to write this expert system., but it became apparent that the factors that influenced the complexity, such as parts count, type of circuit, etc., could be used directly to generate a more satisfactory work-plan estimate. The resulting work plan could be used both as a tool for improving discipline in the engineering process and in reducing the cost of circuit development.

Delphi Method

To overcome the credibility problem of an estimate based on the knowledge of a single expert, it was decided to gather expert knowledge by the Delphi method developed in 1964 by the RAND Corporation (4, 5).

In the Delphi method, a panel of experts is formed and anonymously asked their expert opinion on some matter. For example, what are the engineer and technician hours for developing a given circuit.

The director of the Delphi experiment provides the statistics of the first round of estimates to the panel in terms of median, quartiles, and extremes, along with a summary of the reasons given by the experts in the low and high quartiles for justifying their low and high estimates. The panel then reconsiders their estimate in terms of the statistics and arguments and amends their estimate accordingly, including further justification. The rounds are repeated until convergence is reached or the experiment is terminated for other reasons (such as a badly phrased question).

Since the panel members never know who is advocating a particular line of reasoning or what the others estimated as individuals, a consensus is the result of rational arguments, not personalities.

The Delphi panel for the initial expert system calibration was recruited from a population of over 40 power electronics circuit designers in three divisions at Rockwell's Anaheim, California complex. A panel of experts consisting of 15 power supply circuit designers with 3 years to 25 years of power supply design experience and over 200 years total power supply design experience was formed to provide the initial calibration of the expert system.

Delphi Experiment Results

In the first Delphi experiment, the same circuit schematic used in the mid 1960's was used. This was a 33 part discrete series regulator, which was used as the normal circuit complexity case. The median of the first round of the 1987 Delphi experiment was only 6 percent lower than the 1969 standard hours for the circuit. This was a rather surprising result, until the subelements were examined and it was realized little had changed in 18 years in the way circuit development is done.

The experts examined the results of the first round and modified their estimates. The results of the second round were 19 percent lower than the first round. The arguments that lowered the round-one estimates came from engineers who had used personal computers as productivity aides in recent designs or who had created a template for doing the design. The actual and perceived productivity improvement due to the increased use of personal computers had the most influence in reducing the round-one median estimate to the round-two median.

This round-two data formed the basis for the initial calibration of the expert system. The Delphi panel will be used to estimate other cases and these cases will be used to continue the development and test of the expert system.

The Delphi experiments provided other interesting information. Good circuit designers are not necessarily good estimators, which is no surprise to their managers, who often increase or decrease the estimates of some designers. Also, the least experienced engineers tended to estimate lower than the more experience engineers. However, the experience engineers clustered more on the extremes than the center. Their experience either taught them short cuts with which they were comfortable or made them cautious about the many pitfalls that surround a new design.

Final Calibration

Although the Delphi panel is a more powerful expert than any individual expert (my expert with over 200 years of experience is better than your expert with a limited lifetime of experience), it still needs to be calibrated against accounting data on actual programs to be fully accepted as a valid estimating tool.

This can initially be done by estimating past programs using the aid and calibrating it against actual accounting data from historical records. As the estimating aid is used, a running calibration can be maintained against tasks and programs as they are completed. By controlling the version of the expert system and documenting the rationale for changes, an audit trail can be established.

Intended Uses

The nominal use of the circuit development estimating aid is to aid in preparing cost estimates. Other benefits are less obvious but may ultimately be more important. These include training, engineering discipline, and cost reduction.

Estimating and Scheduling

The most obvious use of the expert system is to increase the value (by the use of the work-plan method) and reduce the time (by the use of a computer aid) in preparing cost proposals for the development of power supply circuits. When compared to present estimating methods (not the work plan method), a factor of two productivity improvement is expected. More important, an estimate based on a detailed work plan, now usually unaffordable, will be produced.

Through the use of the Delphi panel of experts and calibration to historical cost records, the credibility of the estimate should also improve.

Training

Most engineering new graduates have the necessary theoretical and analytical skills to quickly learn the process of engineering a product in a particular industrial environment, but they still must learn the final process through company training programs or on-the-job training. The detailed descriptions of the activities and outputs of each task of the process that is embedded in the estimating aid can help with this training function.

Engineering Discipline

The critical importance of returning to a disciplined engineering process is the central message in two recent government documents (6, 7). These documents identify critical areas in the management, engineering, and production process that can jeopardize a program. The documents provide remedies in the form of discussion, templates, and checklists.

An expert system can be viewed as a template similar to those describe in the government documents. The expert system, when written with this purpose in mind, can serve as a repository of lessons learned, a template of how to do the task in a disciplined way, and a management checklist to insure the process is under control.

For example, the estimating aid expert system collect the subelements that apply to a specific design and places them is schedule order. By printing these subelements, the engineer is given a list of activities and required outputs for his specific design that keeps him on a disciplined track. The same list can be used as a precisely tailored management checklist at periodic reviews of the design effort. Remedies for any failures of the process can be incorporated in the expert system so that future designs benefit from the lessons learned.

Cost Reduction

The work-plan method of producing an estimate has a decided advantage over other methods in that it gives visibility into the location of the high cost items and the details of what makes them high cost. When the work plan is embodied in an expert system the expert system becomes a "what if" tool that can be used to explore more efficient ways of doing things.

Cost reductions can result by exploring methods to improve the efficiency of performing each activity, combining activities in a more synergistic way, and by eliminating subelements not essential to the design process for a specific contract.

In the cost-reduction application, the expert system is modified to reflect the new design environment. The unmodified and modified expert systems are then used to generate estimates for various types of circuits. In this manner, credible cost data can be produced on the productivity improvements that can be expected from various investment strategies.

For example, the effect of using personal computer work stations to increase the productivity of analysis, and decrease the need for breadboarding, can quickly be evaluated by modifying the expert system to reflect the new design environment.

Future Directions

The development of the actual circuit is only one of the design activities necessary to create a product. Performing trade studies, writing test specifications, fabricating and testing engineering and prototype modules, etc., are all activities that need to be estimated. The domain of the estimating aid can be expanded to include these activities or separate expert systems can be developed.

The estimating aid is based on the work being performed in a specific environment with a specific skill level of engineer. For other environments and skill levels, the output of the expert system has to be modified by a learning curve. An expert system to generate this learning curve and to modify the output of the estimating aid is a desirable enhancement.

The estimating aid can be expanded to other analog circuits and a similar aid could be written for digital circuit development.

CONCLUSIONS

A power supply circuit development estimating aid has been developed using an expert-system shell. By using a detailed work-plan method for generating the estimate, the estimating aid has other beneficial applications. These include training in the power supply circuit development process, providing a template for a disciplined approach to power supply circuit development, and producing a tool for exploring methods of reducing the cost of power supply circuit development.

REFERENCES

(1) P. Harmon and D. King, "Expert Systems Artificial Intelligence in Business". John Wiley &Sons, New York, 1985.

(2) J. Foutz, "The Use of Expert System Shells for Developing Design Aids for the Power Supply Designer", Proceedings of Power CAD '87 , Los Angeles, CA, September 10, 1987, Darnell Research Inc., Garden Grove, CA.

(3) EXSYS, Expert System Development Software, EXSYS, Inc., P.O. Box 75258, Contr. Sta. 14, Albuquerque, NM 87194.

(4) J.P. Martino, "Technology Forecasting for Decision Making", American Elsevier Publishing Company Inc. New York, 1972. Chapter 2, Delphi.

(5) H. A. Linstone and M. Turoff (eds.). "The Delphi Method, Techniques and Applications", Addison Wesley Publishing Company, Reading, Massachusetts, 1975.

(6) Department of Defense Manual 4245.7-M. "Solving the Risk Equation in Transitioning From Development to Production, Critical Path Templates + Management Decisions Based on Assessment of Risk = Low Risk Transition", September 1985.

(7) Department of the Navy NAVSO P-6071, "Best Practices, How to Avoid Surprises in the World's Most Complicated Technical Process, The Transition from Development to Production", March 1986.


Copyright Notice. The following copyright notices must be displayed on the initial screen displaying IEEE-copyrighted material electronically:

"© 1988 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE."

"This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each authorís copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder."

For additional copyright information see http://www.ieee.org/about/documentation/copyright/policies.htm.

Original: Foutz, J., Power Supply Circuit Development Estimating Aid - An Expert System Application, IEEE Applied Power Electronics Conference Record, New Orleans, February 1-5, 1988, pp. 64-71. Revised for Web publication with no changes in content 20 February 2001. Revised 27 February 2001


You can find an expanded list of power supply circuit design subelements at Power Supply Circuit Design Subelement Task Descriptions.

Other papers by Jerrold Foutz

Author: Jerrold Foutz, foutz@smpstech.com