session 20 - km for productivity
TRANSCRIPT
-
8/8/2019 Session 20 - KM for Productivity
1/17
Promoting Productivity in Manufacturing Companiesin Developing Countries: An Information System forManaging and Querying Knowledge Bases in theAutomotive Industry in Mexico
Andres G omez de Silva GarzaDepartamento Academico de Computacion, Instituto Tecnologico Autonomo de
M exico (ITAM), Ro Hondo #1, Colonia Tizap anSan Angel, 01000Mexico, D.F.,
M exico. E-mail: [email protected] Lidia Franzoni Vel azquezDepartamento Academico de Computacion, Instituto Tecnologico Autonomo de
M exico (ITAM), Ro Hondo #1, Colonia Tizap anSan Angel, 01000Mexico, D.F.,
M exico. E-mail: [email protected]
Vctor Cruz MoralesDepartamento Academico de Ingeniera Industrial y Operaciones, Instituto
Tecnologico Autonomo de M exico (ITAM), Ro Hondo #1, Colonia Tizap anSanAngel, 01000Mexico, D.F., Mexico. E-mail: [email protected]
ABSTRACT
The purpose of this article is to demonstrate that a low-budget information system using knowledgebases can be developed and adopted in a manufacturing firm to improve the productivity of theorganization. The automotive industry uses complex software systems to perform product engineeringat all stages of the production process, from conceptual design to manufacture. For each of thesestages, different software tools of great complexity are employed, and the people using these tools needconstant support in order to continue being productive even when problems arise with the softwarethey are using. Given the amount and variety of queries made by product engineering personnel tothe software support staff, said staff can benefit from a software tool that stores their expertise forreuse and can be queried in order to generate quick solutions to problems with product engineering
software. In this article, we describe an information system that uses these ideas related to case-basedreasoning in order to flexibly and efficiently reply when problem situations are encountered by theproduct engineering staff of a major automobile manufacturer. This setup can relieve some of theburden placed on the software support staff and reduce their response time. The information systemresulted from a collaboration between a large multinational company and our university, specificallyits undergraduate computer engineering students, benefitting all parties involved in the project andhelping in the development of our country, Mexico. C 2007Wiley Periodicals, Inc.
Keywords: product engineering; case memory; knowledge base; CAD
Information Technology for Development, Vol. 13 (3) 253268 (2007) C 2007Wiley Periodicals, Inc.
Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/itdj.20073
-
8/8/2019 Session 20 - KM for Productivity
2/17
-
8/8/2019 Session 20 - KM for Productivity
3/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 255
Even experienced designers (let alone intermediate- or beginner-level designers) require
technical software support to clear up issues or solve problems that arise during their use of
CAD tools. Even though the use of this type of software tool can be quite intuitive (given
enough experience with them), the results can be highly dependent on the path followed
during the design process. There is no one standardized design process that can always befollowed, and many problems with CAD tools arise out of subtle interactions between the
order in which different substeps were performed, meaning that in similar circumstances,
the same problems may not arise. Some problems may occur directly due to bugs that may be
present in new releases of the software, and patches created by the software developers must
be applied to the tools for the problems to be eliminated; designers using the software tools
are not necessarily aware of the existence of these bugs and their patches. CAD software
also depends greatly on the hardware platform on which it runs, and some problems that
designers have to face might be due to problems with the size of real or virtual memory,
the performance of the graphics processor, and other such hardware-related issues. All
these types of problems can manifest themselves in different ways, and their detection (and
solution) by someone who hasnt had the chance to face them often enough to develop
some expertise may not be at all easy or obvious.
The staff that gives support to CAD tool users in an enterprise involving design must be
highly experienced and specialized in order to be able to provide rapid solutions to these (or
any other) types of problem so that the users can continue with their design-related activities
with a minimum of delay or interruption. However, not everyone providing software support
can be expected to have the same amount or the same type of prior experience. Some may
have encountered only certain variants of a particular problem or may not remember (i.e.,
may not have mentally classified in an adequate fashion) all the possible variants associated
with a given issue. Many may not have any experience at all with recently released versions
of the software package. Finally, support staff will not always be available to try to addressa given query, and even when they are, they cant be expected to be able to respond
immediately.
These problems that may arise in a company that relies on CAD software leads us to
the opportunity to suggest possible solutions. One possible solutionthe one we will be
focusing on in the rest of this articleis for the company to have an information system
(with its associated knowledge base) available both to the support staff and to the users
of CAD software. The information system would serve multiple functions. It would have
the capacity to systematically store new knowledge related to the use of the CAD tools
as said knowledge accumulates. Both users and support staff would have an interface to
the system in order to perform queries and access the knowledge in it. The knowledgebase would store each problem/issue encountered by designers as well as the solutions
eventually found by the support staff for said problems. The knowledge base would serve
as a way for the support staff to share and distribute their expertise among themselves and
with the CAD system users. It would also become an historical, time-stamped document
that traces the different problems encountered and the solutions that have been found for
them (as well as who has encountered the problems and who has provided solutions for
them). Finally, it could be used as a basis for training new advanced users of the CAD tools
or new support staff. It is important to distinguish this type of system from the online or
printed documentation that may be available for the CAD software because it organizes the
information in a different, complementary fashion. It can also describe problems and their
solutions at a higher level of abstraction than the information on typical problems and
known bugs normally found in the documentation or Web sites associated to software, and
-
8/8/2019 Session 20 - KM for Productivity
4/17
256 GOMEZ DE SILVA GARZA, FRANZONI VELAZQUEZ, AND CRUZ MORALES
it can store, as one unit in the knowledge base, problems that arise due to the interaction of
multiple factors together with their solutions. In summary, an information system of this
type embodies, in a centralized repository of knowledge, a subset (in this case the subset
relating to software support for CAD tools) of corporate expertise that would otherwise be
distributed among the brains of all the employees related to this task within a company andaccess to which would not be available in an automated manner.
2. RELATED WORK
As several researchers have remarked, the automobile industry is one of the areas of
human endeavor that has most been affected by the introduction of industrial software
over the past few decades, and yet there are still many challenges and many areas of the
industry in which further improvements can be made as far as providing software support
(Grimm, 2003). Expert systems and knowledge-based systems (Giarratano & Riley, 2004)
have been employed in various tasks related to designing, manufacturing, and operatingautomobiles. For instance, Anonymous (1998) describes a rule-based system used by a
major car manufacturer during the production process to find and repair malfunctions in
their printed circuits; and Sachenbacher, Struss, and Carlen (2000) present a prototype of
a system that is set in a car and diagnoses its operation in real time based on a model
of the cars mechanical systems. On the other hand, the type of tool we describe in this
article is not an autonomous reasoning system but rather just a large knowledge base and
the software needed to administer it. As such, it fits into the area of information science
normally classified as knowledge management (Frappaolo, 2006); yet, our contribution is
that it is usually corporate knowledge of the administrative type (e.g., on business processes)
that is handled by this sort of tool rather than expertise on problems related to the use ofCAD software. Central to knowledge management is the issue of how knowledge is indexed
(organized), and even though this has been an area of interest for a long time now, the past
decade has brought on a flurry of activity due to the ever-increasing use of the World Wide
Web for information storage, search, and retrieval, and the importance of Web searching
strategies. Kim and Seo (2002) did some recent work on indexing (like our system) that uses
natural language words as indices to stored knowledge. On the other hand, we dont just
use natural language words as indices, we combine natural language with more structured
indices in our system, as described belowan approach that is also espoused by Jones,
deBessonet, and Kundu (1988).
3. AUTOMOBILE INDUSTRY CASE
3.1 Problem Scenario
In the rest of this article, we describe an information system with the characteristics
discussed in section 1. For reasons of confidentiality, we shall refer to the company for
which we built the system as Car Manufacturer (CM). This company is the Mexican affiliate
of a multinational automobile manufacturer, and it uses a CAD software package that we
shall call CAD System. The main goal of our system, called CADHelp, is to provide a
knowledge base that will contain the problems and solutions found by users of the CAD
Systems CAD platform. In CM, when the designers using the CAD System encounter a
problem, the procedure prior to the introduction of CADHelp was as follows:
-
8/8/2019 Session 20 - KM for Productivity
5/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 257
1. The person with the problem generally filled out an online form in which the problem
was described in multiple ways, both by selecting predefined options from sets of
radio buttons and by filling out a text box with a natural-language description of the
problem.
2. The people in the technical support group who received the online query were incharge of figuring out the root cause of the problem, usually by trying to repeat the
conditions which led to the problem in their own computer, and then contacting the
CAD System user that originally reported the problem in order to suggest a solution.
The knowledge base in CADHelp is intended to help the technical support staff by
providing them access to previously encountered problems and their solutions, such that if
these problems reoccur, a response can be offered to the designer that encountered them
more quickly than the first time they happened. In addition, if one person in the software
support group has already solved a problem, once he or she enters the information related
to the problem and its solution in the system, the knowledge base in CADHelp representsshared corporate knowledge: All other software support personnel will have access to the
solution found by their colleague and will be able to provide a quick response to the CAD
System user that reported the problem rather than having to find the solution to the problem
again.
3.2 Proposed Solution
Artificial Intelligence (AI) is a complex multidisciplinary field; but, one AI technology that
immediately comes to mind that has the characteristics needed to implement a system that
can achieve the goal of the project is Case-Based Reasoning (CBR; Bergmann et al., 2003;Kolodner, 1993; Leake, 1996; Watson, 1997). CBR arose as an AI methodology in the late
1980s and early 1990s as an alternative to traditional expert systems, which used other
methods or strategies to perform their assigned task. In this project, we didnt propose a
CBR system because CADHelp does not reason by itself; it is not an autonomous system,
as it simply provides its users with expert knowledge that the user then has to employ
in order to solve problems (reason) more effectively and efficiently. However, the entire
internal architecture of CADHelp is based on having cases (this is the type of knowledge
that it stores and provides to its users), and the main characteristic of a case-based system
is that its design is based on the existence and use of past experiences or precedents, so
CADHelp is a case-based system even if it is not a case-based reasoner.These experiences or cases are stored in a case base (also known as a case memory). Each
case stored in memory contains all the relevant information about a specific experience in
the domain of the system (in this case, problems that arise when using CAD System) that
might help in similar situations in the future. This relevant information must include at
least a description of a known problem that was encountered in the past and a description
of the solution that was eventually found for that problem (but sometimes can include
additional related data or annotations if it is deemed to be helpful). This is exactly the type
of information that CM has on previous episodes of problems with the use of CAD System;
however, CM has the information distributed among the brains of their support staff and, to
a much lesser degree, on paper (in a disorganized, unstructured fashion). This is why CBR
came to mind as a technological solution for building CADHelp. Figure 1 shows the case
memory in CADHelp.
-
8/8/2019 Session 20 - KM for Productivity
6/17
258 GOMEZ DE SILVA GARZA, FRANZONI VELAZQUEZ, AND CRUZ MORALES
Figure 1 CADHelps case memory.
Once a case memory has been built, the next step is to determine how to retrieve thecorrect casesthe most relevant onesgiven a new problem. If there are only three or four
cases in memory, then each one can be compared to the description of the new problem
to determine the degree of similarity or relevance and the best one(s) retrieved. However,
when there are many cases in memory (which is the situation in CADHelp), case retrieval is
more efficient if the cases that are not similar or relevant to the new problem can be skipped,
ignored, or discarded from the list of potential solutions without even considering them
(i.e., without comparing them to the current situation) in order to determine their relevance.
This implies the need to come up with an internal organization for the case memory rather
than to have a flat one (consisting simply of a list of cases).
The problem of deciding how to organize the case memory internally is related to theproblem of deciding how the information in the description of a new problem is going
to be used to decide which case(s) to retrieve from the memory, something known in the
CBR literature as the indexing problem. Just as a library has a catalog in which a person
can search for a given book based on multiple criteriausually at least author(s), title,
and main topic(s), a flexible indexing scheme results in a case memory organization that
supports flexible and intelligent case retrieval. The reason that it is desirable to have flexible
and intelligent case retrieval is that the retrieved case(s) is where the potential solution(s)
to the new problem, (which needs to be presented to the user) is stored. Thus, the overall
effectiveness of a case-based system depends on the results of this case retrieval process
producing the best results possible.
In order to decide how we would index the cases in CADHelp, we observed the ways in
which the users of CAD System in CM currently describe their problems using the online
-
8/8/2019 Session 20 - KM for Productivity
7/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 259
form provided by CMs software support personnel. We also observed some descriptions
of old problems that had been written down on paper (not in a very systematic manner
nor covering a large period in CMs existence). Based on these observations, we noted the
terms and concepts used in CM to describe the problems encountered by CAD System
users and the way in which these terms and concepts are interrelated, something knownin AI as the ontology of the domain being studied. This indexing method is similar to
that suggested by (LaBrie & St. Louis, 2003), although in that paper, it is MIS Quarterly
articles that are stored in their knowledge base, not solutions to problems encountered in
the use of software packages.
As a result of this analysis, the indexing scheme we propose for CADHelp is as follows:
CAD System problems can be classified into problems arising from software is-
sues (e.g., the file that contains a particular design has been corrupted or cannot
be read/written by the package); more general conceptual design issues (e.g., the
user does not know what a blend is); or interface-related issues (e.g., problems
with making multiple visual layers of a design compatible). One of these options
is currently selected by the user when filling out the online problem report form.
The chosen option can be used as an index, where each of the different options
points to a different subset of the cases in memory. If it is decided to classify CAD
System problems according to other criteria, then those criteria can also be used
as additional, parallel indices to point to different specific subsets of the cases in
memory.
A natural-language description of the problem is also given currently by the user when
filling out the online problem report form. A search can be made in this description for
certain keywords that carry a high semantic content (Russell & Norvig, 2002). Words
like articles (e.g., a and the) and prepositions (e.g., from and for) do not carryany information that can be used to discriminate between one case or another. On
the other hand, words like nouns (e.g., file and object), verbs (e.g., save and
render), and adjectives (e.g., invisible and large) can be used as indices to point
to subsets of cases that have relevant content (because, for instance, not all cases have
information related to problems with a file).
Given the description of a new problem, then, some cases in memory will not be
retrieved at all, some will be retrieved due to their match with a few indices, and some
will be retrieved due to their match with many indices. The retrieved cases can be
ranked according to how many indices pointed to them (the implementation does
not necessarily involve pointers in the computational sense, but conceptually it is thesame phenomenon which is taking place). The more popular case is the one most
likely to contain the required solution to the new problem (but the rest can be kept and
made available to the user as alternative solutions to the new problem, maintaining
their prioritized ranking according to the number of matches).
This indexing scheme is flexible and intelligent because
given a problem description, it does not provide only one solution (which may or
may not turn out to be correct once it is examined by a human user) but rather many
alternative solutions ordered according to their likelihood of relevance;
new criteria can be incorporated as indices that help discriminate even further between
the different retrieved cases as more and more knowledge is acquired; and
-
8/8/2019 Session 20 - KM for Productivity
8/17
260 GOMEZ DE SILVA GARZA, FRANZONI VELAZQUEZ, AND CRUZ MORALES
the case memory ends up being partitioned into overlapping subsets according to the
different parallel criteria used as indices (which is why there are multiple ways to
arrive at a given case).
Incorporating new cases (or deleting old ones) involves very few changes to the memorystructure (each index in the table of keywords points at a list of cases, and this list has to
grow or decrease by one if the added or deleted case involves the given keyword), making
this sort of learning highly efficient rather than involving a massive reorganization of the
case memory and its contents.
Figure 2 illustrates the indexing scheme used for CADHelps case memory. As can be
seen in the figure, some indices point to several cases, some to just one, and some to none
at all, illustrating the flexibility of the indexing scheme. These ideas result in a dynamic
system whose knowledge base can be easily modified. As can be seen, this solution has
been designed to be quite generalit can be easily modified to incorporate help on other
software packages, not just CAD System, to permit English queries, not just Spanish, etc.
It was thought up to be scalable, with the idea of expanding its contents and capabilities
without having to alter its structure or design.
CADHelp is not the first case-based system used in the automobile industry, and here we
cite just three examples. One example is the project described in Zeid, Gupta, and Bardasz
(1997) in which CBR is used to generate sequences of actions needed to disassemble
cars and trucks in order to recycle their parts. Another example is the system discussed
in Struss and Price (2003), which performs real-time diagnosis of automobile systems as
the car is being used, and then operates the vehicle using model-based reasoning, where
the models are based on generalizing multiple specific experiences (cases). Finally, a third
Figure 2 CADHelps indexing scheme.
-
8/8/2019 Session 20 - KM for Productivity
9/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 261
example is the work of Brueckner and Gerth (2005), which proposes the use of multiple
AI techniques, including CBR, in the implementation of a system that designs the chassis
of an automobile according to the requirements specified by the user. As can be seen,
these systems serve a wide variety of functions, all of which are different from CADHelps
purpose of storing and recovering software support knowledge. This type of system, knownas helpdesk systems, has also previously been studied from the point of view of CBR
(though not specifically geared towards the automobile industry). For example, Kai, Raman,
Carlisle, and Cross (1996) propose that the cases (and their indices) in a case-based helpdesk
system store only free-form text (in natural language).
However, in CADHelp we propose a slightly different approach in which only certain
keywords extracted from natural language text-words that carry information on the meaning
and context of the problem associated with the case-should be used as indices, not entire
sentences or phrases, and certainly not all words that may be present, in any combination,
in the natural language description of the problem. In addition, in CADHelp we complement
the indices obtained from natural language words with more structured information. This
additional information can be input by the user through the selection of different options
via radio buttons in the interface, such as a classification of the type of problem involved
in each case.
4. ANALYSIS OF THE AUTOMOBILE INDUSTRY CASE
4.1 Implementation and Design Methodology
One of the most well-known life-cycle models of the software development process is the
waterfall model first mentioned in Royce (1970). This model consists of seven phases:problem definition, requirements analysis, design, implementation, testing (validation),
integration, and maintenance, as shown in Figure 3. The seven phases describe the steps
through which a software system passes from the beginning of its development to its release
and continued operation. The waterfall model describes the methodology we followed in the
development of CADHelp, though our participation in the project ended upon completion
of the testing/validation phase.
Prior to implementing CADHelp, we conducted a series of interviews with CM employ-
ees (designers and software support staff) in order to define and understand the problem
clearly. While only the CM executives, employees, and the three us were present at the
initial stage of this problem definition phase, undergraduate students soon joined the in-terviews and continued on the project through the rest of the phases. Having finished the
initial problem definition phase, we proceeded to the requirements analysis and design
phases, taking into account the constraints placed on us by CM. In this phase, the three
authors of this article provided most of the guidance as far as choosing which approaches
and technologies to use to tackle the problem, but students ended up providing a lot of
input too. Part of this guideance included focusing on information retrieval and knowledge
management rather than thinking about developing an expert system, as CMs executives
originally requested (probably as a result of misunderstanding the fact that this term im-
plies autonomous reasoning or problem solving capabilities). Finally, we implemented
CADHelp using Excel to store the cases and other important knowledge and using Visual
Basic for Applications (part of Excel) to design the interface and program the different
search and retrieval functions needed when a user queries the system. This implementation
-
8/8/2019 Session 20 - KM for Productivity
10/17
262 GOMEZ DE SILVA GARZA, FRANZONI VELAZQUEZ, AND CRUZ MORALES
Figure 3 Waterfall model of software development.
phase of the project and the subsequent validation phase were completely performed by the
students.
The policy makers at the corporate headquarters of CM imposed the use of these particular
platforms (Excel and Visual Basic for Applications) for storing the knowledge base and
developing the system. For standardization purposes, there are constraints on the software
that the subsidiaries of CM throughout the world must use to perform certain tasks (forexample, using spreadsheets or programming) or to develop their own systems. These
constraints depend on the software providers that CM uses and the software licenses they
already have or are willing to buy or negotiate. Thus, it wasnt feasible in the CADHelp
project to use dedicated knowledge management tools or little-used programming languages
such as those commonly employed in AI projects. On the other hand, the policy makers
at the CM corporate headquarters have not identified the development and storage of
knowledge bases as a task that is important to standardize or even to carry out, so they let
their subsidiaries decide whether they are going to do it (corporate policy at CM dictates
that most operative problems be dealt with independently by each subsidiary). And if they
decide to do it, it is up to the subsidiaries to figure out how to do so. In this particularsituation, the Mexican subsidiary of CM decided that they would like to have a system
with HelpCADs characteristics, and they carried out the project by getting in touch with
our university as part of an agreement of collaboration that CM has with several Mexican
universities (something which, as mentioned earlier, has historically been uncommon in
Mexico). There has been talk about the possibility for other subsidiaries of CM (in other
countries) or even the people in their corporate headquarters to begin using CADHelp but
this has not yet happened.
Returning to the characteristics of the system, in CADHelp there are two possible types
of users: regular users who can describe new problems that they encounter as they use
CAD System and expert users who have administrative priviledges and can modify the
information stored in the knowledge base and the database of authorized users. Even more
importantly, expert users are those that can check to see if any new problems (previously
-
8/8/2019 Session 20 - KM for Productivity
11/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 263
entered by regular users) are pending, find solutions to them, and add these solutions (and
related problem descriptions) to the case memory to make them immediately available
for reuse if the same problems occur again in the future. The use of Visual Basic for
Applications for the implementation means that HelpCAD has a friendly, easy to use
interface, and therefore, we avoided having the complicated and confusing interfaces thatsoftware packages that manage knowledge bases normally have.
4.2 Knowledge Base Interface
In CADHelp, access to the knowledge base is through one of two interfaces; which one
is displayed is based on the type of account with which a user has logged on (regular or
expert). Figure 4 shows the multiple functions available through each of the interfaces for
the two types of user.
Figure 5 shows the options available in CADHelps main menu to expert users. As can
be seen, it is this type of user that administers the cases stored in the systems knowledgebase.
The knowledge base in CADHelp can grow without limits as more and more knowledge
is acquired and input by the experts. The accumulated knowledge is organized and indexed
according to multiple criteria, including the versions currently being used of the CAD
System. The multiple indexes allow for a rapid retrieval of the relevant information each
time the knowledge base is queried. The indexes used reflect the type of information that
the experts would use naturally to describe and classify CAD System problems even if they
didnt have an information system such as CADHelp to rely on for the storage and retrieval
Figure 4 Functions available to the two types of user in CADHelp.
-
8/8/2019 Session 20 - KM for Productivity
12/17
264 GOMEZ DE SILVA GARZA, FRANZONI VELAZQUEZ, AND CRUZ MORALES
Figure 5 Interface options available to expert users in CADHelp.
of knowledge. Retrieved cases are assigned points according to how many indices extractedfrom a query point to them, thus allowing the system to rank them according to their likely
relevance/usefulness, as shown in Figure 6. The knowledge base embodies the expertise
of the entire CAD System software support staff in CM and makes it easily accessible to
both the support staff and the designers that use the CAD System. This expertise stays with
CM regardless of whether the support staff who originally entered the knowledge is still
with the company, still has the same job/post within the company, or other factors that can
sometimes hamper the transfer of expertise in an enterprise. The amount of time that needs
to be invested in order to fill the knowledge base with some initial cases can quickly be
recovered due to the amount of time that CADHelp saves in the error reporting and error
correction phases of the design process.In CADHelp, regular users can describe new problems that they encounter as they use
CAD System in their daily chores, or they can use CADHelp to look for solutions to
problems that they have previously reported (assuming that a solution to the problem has
already been found and entered by an expert user), as shown in Figure 7. A regular user
can also provide a natural-language (in this case Spanish) description of the problem when
filling out the online report form, and the system can look for keywords contained in the
text in order to retrieve and sort relevant solutions stored in the case base.
5. RESULTS AND CONCLUSIONS
The first thing that must be said is that the answer to our research question is a resounding
yes! It is definitely feasible to promote the productivity of manufacturing countries in
-
8/8/2019 Session 20 - KM for Productivity
13/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 265
Figure 6 Example of results displayed by CADHelp after a Quero.
developing countries through the development and deployment of low-budget informationsystems. The car manufacturer (CM) for which we built the CADHelp system has greatly
benefited from the design and implementation of the system. Their software support staff
used to have a few folders containing brief, unordered, and unsystematically annotated
solutions written on paper to some problems faced by users of their CAD System. The dif-
ficulties inherent in this prior knowledge repository system that did not rely on information
technology can be easily grasped. The CADHelp system has the necessary characteristics
to systematize and automate much of the work done by the software support staff in CMs
Mexican subsidiary, and its development did not require the investment of large amounts of
time, money, or effort. CADHelp was deployed with an empty knowledge base and access
privileges for only one expert user and one regular user. Within weeks, the information that
had previously only existed in written form in a disorganized fashion had been added and
several user accounts, for both expert and normal users, had been created as needed. The
-
8/8/2019 Session 20 - KM for Productivity
14/17
266 GOMEZ DE SILVA GARZA, FRANZONI VELAZQUEZ, AND CRUZ MORALES
Figure 7 Query performed by a regular CADHelp user.
software support staff, whose work from anecdotal accounts has become more efficient and
less tedious, have been systematically entering their experiences into CADHelp by orders
of the engineering design management, which shows not only the usefulness of the system
but also the commitment of CM to the project. The knowledge base as it stands can, at
the same time, collects reports of new problems, incidents, and issues related to the CADSystem and stores (for future retrieval) the solutions for these issues found by the support
staff. The idea is that regular users soon begin employing the system in their daily tasks
as well, minimizing the need for them to rely on the support staff, who can thus focus on
the more difficult novel problems that may arise. Developing CADHelp involved the use
of inexpensive and widespread standard software programming platforms, such as Excel
and Visual Basic for Applications, instead of requiring complex and expensive technology
such as expert system shells.
The solutions we found for CMs needs are quite general and not tied exclusively to the
environment in which the company operates (the automobile industry) or to the specific
task (support for users of CAD packages) that needed to be performed. This task requiredthe corporate expertise of the software support personnel to be stored, managed, and
retrieved in a more systematic, automated manner. The CADHelp systems interface can be
easily adapted to other applications that require knowledge bases accessed through multiple
indices that represent the global expertise of an organization. The indexing scheme is also
general and can be employed for organizing cases (or knowledge represented in other
ways) that represent different types of knowledge, not just for software support. The use
of natural language words as partial indices to the cases stored in the knowledge base is
independent of which natural language is employed to describe problems, so portability to
other subsidiaries of CM worldwide is perfectly feasible. One future potential application
is to use the same interface that CADHelp has, with a different knowledge base, to provide
support for users of PLM (Product Life-Cycle Management) systems, which are similar to
CAD software due to their size and complexity.
-
8/8/2019 Session 20 - KM for Productivity
15/17
THE AUTOMOTIVE INDUSTRY IN MEXICO 267
The development of the CADHelp system has been a perfect opportunity for collaboration
between the industrial and academic sectors in order to solve a real-world problem. The
alternatives for CM would have been to hire a team of consultants or to ask their own
employees to develop a system such as CADHelp. Both of these options would have been
too costly for different reasons. The first one is due to its high monetary cost, and the secondbecause it would have implied reassigning software support and software development
personnel from their ordinary tasks to the development of this project for lengthy periods
of time. Thus, in the end, probably nothing would have ended up being developed (and the
original inefficient, unsystematic method of doing things would have remained). Thanks to
the existence of the industry-university collaboration agreement, CM did not have to invest
money or time on this project (with the exceptions of the first brief interviews during the
problem definition and requirements analysis phases of the project and the quick system
tests during the debugging phase of the implementation).
The students who participated in the project, some less than halfway through the course
load associated with their degree, proved to be mature, knowledgeable, and skilled enough to
tackle the software development tasks that were assigned to them. Despite a limited amount
of time (the entire project had to be completed in 4 months, from problem specification
to deployment) and the absence of any monetary reward (neither students nor professors
benefitted economically from participating in the project, though both gained valuable
experience), the project was carried out and completed successfully without a hitch and
within the alloted time. At the same time, the students and professors of our academic
institution had the opportunity to work together on a real-world project outside of the
normal classroom setting, and they were able to learn about CAD tools (and the problems
related to their use) and apply AI techniques and information technology in general to the
management of corporate knowledge.
In other words, the project benefitted all parties involved on both the industrial and theacademic side of things. Perhaps even more beneficial is the snowball effect that is already
taking place within CM: our university is presently working collaboratively on three more
projects. Perhaps, eventually this will lead to other car manufacturers and other sectors of
industry to join forces with academia in research and development projects that are not just
mutually beneficial to the participating parties but will have further reaching consequences
for development as a whole in Mexico.
REFERENCES
Anonymous (1998). Ford adopts rule-based cim strategy. Intelligent Systems Report, 15(5), p. 12.Bergmann, R., Althoff, K.-D., Breen, S., Goker, M., Manago, M., Traphoner, R., et al. (2003).
Developing industrial case-based reasoning applications: The inreca methodology (2nd ed.).Springer Verlag, Lecture Notes in Computer Science, 1612.
Brueckner, S. A., & Gerth, R. (2005). Applying distributed adaptive optimization to digital car bodydevelopment. In S. A., Brueckner, G. Di Marzo Serugendo, A. Karageorgos, & R. Nagpal (Eds.),Engineering self-organizing systems: Methodologies & applications. Springer Verlag, LectureNotes in Computer Science, 3464, pp. 267279.
Frappaolo, C. (2006). Knowledge Management (2nd ed.). Oxford: Capstone Press.Giarratano, J. C., & Riley, G. D. (2004). Expert systems: Principles and programming (4th ed.).
Course Technology.Grimm, K. (2003). Software technology in an automotive companymajor challenges. Proceedings
of the 25th International Conference on Software Engineering, IEEE Press (pp. 498503).Jones, L. P., deBessonet, C., & Kundu, S. (1988). ALLOY: An amalgamation of expert, linguis-tic and statistical indexing methods. Proceedings of the 11th Annual ACM SIGIR Conference
-
8/8/2019 Session 20 - KM for Productivity
16/17
-
8/8/2019 Session 20 - KM for Productivity
17/17