ABSTRACT

In this paper, we present a method for searching examples for given problem solving heuristics in the electronic patent database and the statistics of our search results. Instead of conducting our search from patent classes to the examples (top-down approach), we have chosen to search solely by keywords derived from heuristics (bottom-up approach). The keywords have been derived from a subset of Polovinkin's heuristics. The examples (patents) found cover almost all fields of technology (as they defined by International Patent Classification). Information about the most preferable classes of heuristics for each technical field is also presented.

INTRODUCTION

Current TRIZ is a mostly empirical methodology. Its heuristics (e.g. Inventive Principles, Trends of Technical Evolution, and Standards) were extracted from many information sources, especially from patents [1, 2]. It is to be noted that TRIZ heuristics, although mostly generic - and therefore lasting - were derived from a subset of the patent database, corresponding to mostly Russian patents available at the time when the heuristics where obtained (approximately from 1960 to 1985). Patents are issued in most countries around the globe and the number of patents constantly grows in each country. This could mean that the current set of TRIZ heuristics is not complete and certainly means that traditional examples are outdated. It is possible to argue that the examples, which confirm or deny the current set of heuristics, are important for TRIZ research [3].

Three factors slow down development of TRIZ through patent research. One is the fact that so-called high-level patents should be considered, and these represent only about a quarter of the whole patent database [1, 2]. The other is the fast expansion of the worldwide patent databases. The third factor is the relatively long period of time that is necessary for TRIZ analysis [4] of a single patent and the absence of a good computer aid for such research. It was proposed [5] to improve effectiveness in research of new TRIZ heuristics by filtering out a large amount of low-level patents using computers and the electronic patent database.

In this paper, we search examples for a set of 100+ heuristics for problem solving and design selected by A. I. Polovinkin from best professional practice of mechanics and machinists in the former USSR [6, 7]. We use the International Patent Classification to evaluate our results. Current (seventh) edition of the International Patent Classification (IPC7) divides technology into 8 sections with nearly 69,000 subdivisions congregated into 120 patent classes, which are further classified into 22 groups of technology [8]. Each IPC7 class has the symbol of an initial Latin letter followed by two Arabic numerals (e.g. A01 and F41). The initial letter represents the IPC7 section that the class or subclass belongs to. The groups in IPC7 have no symbols, but they have titles and definitions. We perform statistical analysis of our results as well as appraise the 100+ heuristics and heuristics classes according to their relative versatility in correlation with IPC7.

PATENT SEARCH

In collecting examples of 100+ Polovinkin's heuristics [3], we have set the goal of patent search to finding at least three high-level patents from different fields of technology for each heuristic. To estimate how many patents should be reviewed, we have used Altshuller's statistics of high-level patents (about 23%) [1] and referred to the precision ratios (about 60%~75%) [9, 10] reported from other searching of electronic bibliographic databases. Therefore, about twenty pre-selected patents should be examined in order to find three high-level patent examples for a given heuristic [3].

Two portals for electronic patent database have been used in our study. One is the US Patent and Trademark Office (USPTO), the other is Delphion Inc (the former IBM patent WWW site). USPTO provides free searching of full-text patents since 1976. Delphion provides free searching of front-page information since 1971, and offers relevancy ranking for the search results. Although the search languages of USPTO and Delphion databases are quite different, both retrieve the relevant patents with the non-adequate patents that are not important for our studies. Because of the portals limitation, we have searched only relatively new patents, issued in the last three decades.

Search strategy depends very much on the purpose of the search and the database being used [10-12]. Our research has used expanded keyword search as the primary strategy. We have taken the following steps to conduct the patent search:

Step 1	Grasp the essence of the heuristic under verification, by formulating its primary function and associated characteristics.
Step 2	Choose keywords from heuristic description as primary search terms. Represent all possible synonyms as well as more specific or more general terms. Formulate the search query.
Step 3	Perform searching on the selected electronic patent database. The result contains relevant and non-relevant patents.
Step 4	Pre-select patents relevant to the heuristic being verified.
Step 5	Study relevant patents. Use basic flowchart of patent investigation [2, 4] to inspect if patents belong to the high-level class.

OVERVIEW OF SEARCH RESULTS

We have collected 586 U.S. patents to illustrate 100+ Polovinkin's heuristics [3, 6, 7, 2] for system transformations. Although we did not purposefully try to expand our search geographically, collected examples include inventions originated in more than 30 countries from around the world. The most frequent countries found are United States of America (59.4%), Japan (16.6%) and Germany (6.8%), and they cover more than 80% of our patent search results. This fact illustrates that Russian-based 100+ heuristics are culture-independent and, most important, are widely used by world's leading industrialized countries.

For each heuristic, we have tried to find as diverse fields of technology as possible. On average, each heuristic has about five U.S. patents as examples. Hence we can presume that each heuristic may cover a few different technical fields. We did not try to accumulate the exhausted collection of examples for any of 100+ heuristics. To see how broad the technical fields covered by search results of 100+ heuristics are, we have analyzed selected patents according to IPC7. Table 1 summarizes this analysis.

The correlation between the number of patents cited for 100+ heuristics [3] and the number of patents in USPTO database is 0.86. This high correlation reflects the purposefully extended area of search by our expanded keyword search described above, and the ranking of the most relevant patents provided by the electronic database. Table 1 also shows that the percentages of citation for Sections C (chemistry and metallurgy) and D (textiles and paper) are significantly below the average percentage of citation (0.022% or about 4,500 relevant and non-adequate patents per example). Separation of the relevant from non-relevant patents is difficult, because we are unfamiliar with many chemical processes (represented in the majority of Sections C and D patents). Thus, relevant patents of these sections could have been easily rejected as non-adequate and/or their level could have been misestimated. When excluding these Sections, the above correlation reaches the very high value of 0.96. This means that our examples represent whole US patent database extremely well and that 100+ heuristics can be used for problem solving in all major technical fields.

Table 1 - Statistics of search results by IPC7 sections

The statistics of 100+ heuristics search results are further expanded into details according to eight classes of 100+ heuristics, as shown in the Table 2. Almost all heuristic classes have examples for every IPC7 section.

For now, analysis has shown that the results of our patent search widely cover eight major sections of technology. In order to be more precise and to provide additional useful information such as what classes of heuristics for system transformation are the most preferable (or promising) to a inventor's area, we should analyse further the specific technical field. The analysis of examples according to 22 groups of IPC7 classes is summarized in Table 3. Again, we have examined the correlation between the number of patents cited in 100+ heuristics and the number of patents in the groups of IPC7 patent classes that are listed in the Table 3 for the reader convenience. The correlation is 0.88, which illustrates the broad technical fields in our patent search. There is no example found for group B81-B82, because there are only few patents in the USPTO database for this group.

Table 2 - Statistics: IPC7 sections vs. classes of 100+ heuristics

Table 3 - Statistics of search results by groups of IPC7 classes

According to Table 3, we have found that the following eight groups contain the greater part of patents found in the results of our search:

1. Instruments (148 examples)

2. Electricity (105 examples)

3. Transporting (77 examples)

4. Shaping (62 examples)

5. Health and amusement (45 examples)

6. Engineering in general (27 examples)

7. Separating and mixing (21examples)

8. Chemistry (21examples)

Patents of 100+ heuristics in these eight groups represent more than 80% of all patents found for Polovinkin's heuristics. Quite similarly, these eight groups cover nearly 80% of patents in the USPTO full-text electronic database. It should be noted that the 100+ heuristics were originally formulated based on experience of Russian engineers also from the fields of Instruments (G01-G12), Transporting (B60-B68) and Shaping (B21-B32). Familiarity of these technical fields to the searchers and the large amount of respective patents in the database also contribute to this ratio. However, using wide variety of heuristics for system transformations in these fields might also have influenced in the distortion.

Next, we relate 22 groups of technology according to IPC7 with eight classes of 100+ heuristics (see the Table 4). The most related heuristic class (or classes) for each group of technology is highlighted in the Table 4 for the reader´s convenience. For example, "Health and amusement" technology (A61-A63) relates mostly with shape and material transformations.

EVALUATION OF SEARCH RESULTS

In this section, search results are evaluated by measuring the relative versatility (RV) of heuristics and heuristic classes. The RV measure for each heuristic is calculated as:

Here we have used the 22 groups of IPC7 as categories. The average value of RV is 0.74, while the highest value of a heuristic is 1.00 (the maximum possible number), and the lowest one is 0.33 for the found examples. In our study, the average number of examples (including those from patents and those not [3]) for a heuristic is 5.34. Therefore, a heuristic can have examples from typically four different groups of technical fields as defined by IPC7, which proves the diversification of the heuristics and their examples.

After the RV values of all heuristics have been counted, the RV for each class of heuristics is estimated as an average of RV of all underlying heuristics. It is to be noted that RV is a division-dependent measure. In our case the RV depends to a large extent upon the patent classification system (i.e. number of categories). It is known that the IPC7 system has 8 sections, 22 groups, 120 classes, and a large amount of sub-classes [8]. If we estimate RV by IPC7 sections, the average RV of all classes of heuristics would be 0.63. Therefore, we conclude that "finer" classification will likely result in higher RV measures, and vice versa. Thus, in this study, it is representative to use IPC7 groups in calculating RV of heuristics and heuristic classes.

Table 5 shows the RV figures for all heuristic classes. Higher RV means a wider range of applicable fields of technology. It is interesting to note that the values of the average RV for a heuristic (0.74) and the average RV for a class of heuristics (0.73) are almost the same. Right now we do not know if this is due to a mere coincidence, our calculation approach or other underlying reasons (such as the amount of heuristics in each class or IPC7 features).

Table 4 - Statistics: groups of IPC7 classes vs. 100+ heuristics classes

The classes "3. Transformations in space", "8. Quantitative changes" and "2. Structure transformations" have the highest values of RV although the difference from the average RV is statistically non-crucial. The class "4. Transformation in time" has the lowest value of RV and its difference from the average RV is statistically significant. Why "Transformation in time" has the lowest RV? Obviously, time is more important for technological processes than for technical systems [2], while most patents are issued for the latter category of technology [11, 8]. Perhaps time is a much-ignored resource in inventing. It is interesting to note that the class 4 has the smallest amount of heuristics probably because we cannot reverse, accelerate, squeeze one-dimensional time yet.

Table 5 - Measures of relative versatility for 100+ heuristics classes

CONCLUSION

In conclusion, analysis of our search results shows that 100+ heuristic are culture-independent and can be used in various fields of technology. Using only expanded keyword search plus purposefully finding diverse fields of technology in searching patent examples for each heuristic, we have effectively covered a large part of technical fields in IPC7 classification. By correlating 22 groups of technology with eight classes of heuristics, inventors can have some ideas about which types of system transformations are most related to their fields. The patented examples that we provide for the different technical fields [3] should increase the efficiency of heuristics aid for diversified inventors. The main objective of our research - to provide inventors with a useful set of heuristics, along with updated examples of their use - is being achieved.

REFERENCES

1. Altshuller, G. S. Innovation Algorithm. Worcester: Technical Innovation Center, 1999 (1st Russian edition, 1969).

2. Savransky, S. D. Engineering of Creativity: Introduction to TRIZ Methodology of Inventive Problem Solving. CRC Press, 2000.

3. Savransky, S. D., Wei, T. C., de Carvalho, M. A. 121 Heuristics for Inventors. In: Guide for Inventors, Vol. 2. RO-INI, 2002.

4. Savransky, S. D. How to study patents in the framework of TRIZ. In: Proceedings of TRIZCON99. Novi: Altshuller Institute, 1999.

5. de Carvalho M. A., Wei, T. C., Savransky, S. D. Validation of heuristics for systems transformations. In: Proceedings of TRIZCON2001. Altshuller Institute, Woodland Hills, CA, 2001.

6. Polovinkin, A. I. Theory of New Technology Design: Laws of Technical Systems and their Applications. Informelektro: Moscow, 1991 (in Russian).

7. Polovinkin, A. I. The ABC of Engineering Creativity. Mashinostroenie: Moscow, 1988 (in Russian).

8. International Patent Classification. Available at http://www.wipo.int/classifications/

9. Huang, M. H. Studies of the Concept of 'Relevance' in Information Retrieval. Taiwan Students Book Co., 1996. (in Chinese).

10. Jantz, R. Searching Electronic Databases for Information on Soil Remediation: The Interview and the Bibliography. Electronic Green Journal, April 1999. Available at http://egj.lib.uidaho.edu/

11. Pressman, D. Patent It Yourself. 7th edition, Nolo Press, 1999.

12. Canadian Intellectual Property Office. Searching the Patent Literature in the Electronic Age - A Short Course. PATSCAN, June 1998. Available in the www at http://www.library.ubc.ca/patscan/

BIOGRAPHICAL SKETCHES

Tz-Chin Wei is a project manager at Flotrend Corporation in Taipei, Taiwan. Wei received his BS and MS in mechanical engineering from National Taiwan University in 1993 and 1995. His interest in TRIZ began in 1998, and has studied TRIZ in the Virtual TRIZ college since 1999. In 1998-2002 he trained over 300 people in using computer-aided design tools to accelerate their industrial product design. His research is oriented towards engineering, creativity and TRIZ-related issues.

Address: Flotrend Corporation - 3F, 72, Sungteh Road, Taipei, Taiwan 110, R.O.C.

Phone: +886-2-27587668 Fax: +886-2-27589798 Email: peterwei@ms9.hinet.net

Semyon D. Savransky became acquainted with TRIZ in 1980-1981. Semyon D. Savransky is the author of TRIZ books published in USA and Europe, numerous patents and scientific papers. Semyon D. Savransky is one of the leading TRIZ experts who combines diversified experience in engineering, science and pedagogic. He received Ph.D. in physics 1989 in Leningrad (now St. Petersburg, Russia), and his academic background is split between Novgorod State University (Russia), Universidad Pais Vasco (Spain) and New York City University (USA). Currently S. D. Savransky runs innovative projects for high-tech companies in Silicon Valley, California, USA He is the founder of the Research Center in Novgorod State University - NGPI (Russia) and International Company: TRIZ Experts.

Address: TRIZ Experts - 6015 Pepper Tree Court, Newark, CA 94560, USA

Fax: (509) 271-3704 E-mail: TRIZ_SDS@hotmail.com

Marco Aurélio de Carvalho is a B.Sc. Mechanical Engineer and has a M.Sc. in Production Engineering. He has five years experience in Engineering Design at Electrolux, John Deere and NuPES as well as two years experience in Quality Engineering at Bosch. He is currently Assistant Professor of Mechanical Design and Design Methodology at CEFET-PR (Mechanical Engineering Department) and Researcher at NuPES (Concurrent Engineering Research Laboratory) in Curitiba, Brazil. He is a doctoral candidate at UFSC (Federal University of Santa Catarina). His research interests are TRIZ, Engineering Design and Technology Forecasting. Marco is involved with TRIZ since 1997.

Address: CEFET-PR - Av. Sete de Setembro 3165, Curitiba, PR, Brazil

Phone: +55.41.310.4770 Fax: +55.41.310.4753 Email: marco@nupes.cefetpr.br