Modeling of an expert system for an electric

muffle furnace for simple heat treatment

processes
Natraj Mishra1, Parameshwar Sathyanarayan2
1
ITM University, Sector- 23A HUDA, Gurgaon- 122017, Haryana, India
2
Global Academy of Technology, Bangalore, Karnataka, India
1
Email: [email protected]

Abstract- This article describes the modeling of an expert

system for an electric muffle furnace for controlling the B. The Expert System
simple heat treatment processes like annealing and
normalizing in case of plain carbon steels. The heat Expert system is a computer program using expert
treatment process consists of three stages viz. heating, knowledge to attain high levels of performance in a
soaking and cooling. This paper deals with the heating narrow problem area [3]. It provides powerful tools to
stage i.e., it gives a method to design an expert system that deal with complex real-world systems and to aid in the
sets up the required heating temperature for a given plain transfer of knowledge to engineering practice [4]. It
carbon steel. For this purpose, an expert system was consists of a knowledge base which has information
created having a knowledge base containing information about variation of mechanical (BHN) and physical and
regarding carbon percentage, hardness and austenitizing chemical properties (T, % C) with respect to each other
temperature (temperature at which the iron, iron-based
for plain carbon steels. In case of annealing, the user
metal, or steel changes crystal structure from ferrite to
austenite). The expert system was given two inputs viz. gives two inputs viz. %C and BHN. The output is
Carbon percentage and required Brinell Hardness summing temperature which is used to find out Td for
Number on the basis of which temperature set point was the given steel for the muffle furnace. It is based upon a
set for the muffle furnace using fuzzy logic. The fuzzy inference system (FIS).
simulation of this expert system was done on MATLAB/
SIMULINK. The results obtained after simulation were II. MODELLING OF THE EXPERT SYSTEM
compared with the standard results available in a
standard design data book. The two results were found to The expert system takes two inputs viz. %C and
be in good agreement with each other. It was found that required BHN and provides one output i.e. Td.
the expert system decided the correct values of
temperature set point for different types of annealing
processes for plain carbon steels.

Keywords: expert system, electric muffle furnace, plain

carbon steels, annealing processes, SIMULINK

Abbreviations:
Td = desired furnace temperature/ temperature set-point
T = temperature
Fig. 1: The expert system
BHN = Brinell Hardness Number
%C = percentage Carbon
Figure 1 shows the block diagram of the expert system
FIS = Fuzzy Inference System which is a two input and single output system. The user
KB = Knowledge Base has to provide the values of carbon percentage (%C)
sumtemp = summing temperature within the given steel and the hardness to be attributed
to the steel after heat treatment (required BHN). Based
I. INTRODUCTION upon these two inputs, the expert system gives the value
The work presented in this paper is the outcome of of desired furnace temperature (Td) to which the interior
Master Degree thesis of the first author, titled- Design of the muffle furnace has to be raised to. According to
of an Expert System for an Electric Muffle Furnace [1]. Linkens and Chen [5], the main tasks of developing an
An attempt has been made to develop an intelligent expert control system can be grouped into three parts-
database system containing various physical properties model building of the process, construction of an expert
of materials. This database is essential for computer controller and establishment of a user-friendly
simulation [2] of simple heat treatment processes for interface. In the present case, the development of the
plain carbon steels. expert using MATLAB/ SIMULINK was accomplished
in three stages- creation of knowledge base, generation
A. Aim of rule base and generation of membership functions
using Mamdani type FIS. The user friendly interface is
The aim is to design an expert system for an electric provided by the MATLAB.
muffle furnace for optimizing and controlling the
simple heat treatment processes like annealing and
normalizing in case of plain carbon steels.

1
 A. Creation of Knowledge Base Table 1: Heat treatment datasheet for annealing

The KB includes all the information pertaining to the Heat treatment Heated up to % Carbon range
relationship between carbon percentage and type (degree
austenitizing temperature. Figure 2 shows this Celsius)
relationship.
Spheroidizing 700 0.60 to 1.60

Process annealing 550 to 650/ 0.00 to 0.30

700

Full annealing 30 + A3/ A1 Hypoeutectoid (<

0.77 %)

Normalizing 55 + A3/ Acm ≥ 0.4 %

B. Generation of Rule Base

The rule base is generated using information provided
Fig. 2: Graph showing variation of upper critical temperatures (A3 in a standard design data book [8]. A rule-based expert
and Acm) with carbon percentage system contains information obtained from a human
In figure 2, we can see two curves- A3 and Acm. Curve expert, and represents that information in the form of
A3 corresponds to the austenitizing temperature for rules, such as IF-THEN [9]. It is generated for FIS
hypoeutectoid steels (having < 0.77 % C) while curve having two inputs and single output. The rule base is as
Acm corresponds to the austenitizing temperature for follows:
hypereutectoid steels (having > 0.77 % C). The graph is ƒ If BHN is softest and %C is low then sumtemp
obtained by taking reference from Askeland and Phule is process anneal.
[6]. The KB is also created using information regarding ƒ If BHN is softest and %C is not low then
various heat treatment processes suitable for a given sumtemp is spheroidize.
plain carbon steel. Figure 3 shows the austenitizing ƒ If BHN is softer and %C is low then sumtemp
temperatures for different heat treatment processes for is process anneal.
plain carbon steels. ƒ If BHN is softer and %C is not low then
sumtemp is full anneal.
ƒ If BHN is soft and %C is low then sumtemp is
full anneal.
ƒ If BHN is soft and %C is not low then
sumtemp is normalize.

C. Generation of Membership Function

Membership functions characterize the fuzziness in a
fuzzy set- whether the elements in the set are discrete or
continuous- in a graphical form for eventual use in the
mathematical formulisms of fuzzy set theory [10]. The
FIS has two inputs and one output. Table 2 shows the
membership function ranges for input variables and
output variables. The membership functions are made
using Mamdani approach.
Fig. 3: Summary of heat treatments for steels [7].
From figure 3, it can be seen that spheroidizing is
carried out between 700 deg. C and A1 (727 deg. C)
curve for steels having carbon percentage greater than
0.6 %. Similar observations can be obtained for process
annealing, full annealing and normalizing. Table 1
provides the heat treatment datasheet for annealing [6],
[8].

2
 Table 2: Table showing membership function ranges for the input and output variables of expert system

Variable Membership Parameters

Function
(Trapezoidal)
Required BHN Softest -150, (100, 150,
(Input) 200)
Softer (150, 200, 250,
300)
Soft (250, 300, 350,
400)
Percent Low (0, 0.1, 0.3, 0.4)
Carbon
(Input) Medium (0.3, 0.45, 0.65,
0.8)
High (0.7, 0.85, 1.2),
1.6
Summing Process anneal (-200, -180, -
temperature 100, -50)
(ºC) (Output) Spheroidize (-100, -80, -20,
0)
Full anneal (0, 10, 45, 55)
Normalize (45, 55, 65), 75

III. MODELLING IN MATLAB/ SIMULINK

First input- required BHN has three membership
functions- softest (100 BHN to 200 BHN), softer (150 Figure 4 shows the SIMULINK model of the expert
BHN to 300 BHN) and soft (250 BHN to 400 BHN). system. The simulation work was carried out in
The second input- %C has also three membership MATLAB 2007b. The expert system is governed by a
functions- low (0 to 0.4 % carbon), medium (0.3 to 0.8 set of rules that are collected from the design data book
% carbon) and high (0.7 to 1.2 % carbon). The output, and other references. It makes use of Mamdani type FIS
summing temperature has four membership functions- and takes decision regarding the desired furnace
process anneal (-200ºC to -50 ºC), spheroidize (-100ºC temperature.
to 0ºC), full anneal (0ºC to 55 ºC) and normalize (45ºC
to 65ºC). This output is added to the austenitizing
temperature of the given steel (refer Fig. 2) to give the
desired furnace temperature.

250

Required BHN
BHN

s um tem p

%C Desired furnace temperature , Td

Membership Function & Rule Base

0.6
Austenitizing Temperature (A1/ Acm )
Percent Carbon

%C A1/ Acm

Knowledge Base

Fig. 4: SIMULINK model of the expert system.

From figure 4, it can be observed that the input- percent content of the plain carbon steel. The two outputs-
carbon is fed into two blocks viz. membership function sumtemp and austenitizing temperature are added
& rule base block and knowledge base block. The first together to give the final output i.e., the desired furnace
block has another input- required BHN. It gives the temperature (Td). In figure 5, the internal structure of
output- sumtemp. The second block gives austenitizing the membership function & rule base block has been
temperature as the output based upon the carbon shown.

3
 1
BHN
1
sumtemp
Mamdani Fuzzy Inference System
2
%C
Fig. 5: Internal structure of the membership function & rule base block (SIMULINK model).

A. Generation of Membership Functions in

MATLAB
Figure 6 shows the FIS block for finding the summing
temperature which is further used to find out the desired
temperature.

Fig. 6: Fuzzy logic block to find out the summing temperature.

The two leftmost blocks are for the input variables- 1) Membership Functions for Input 1 (Required
required BHN and %C while the rightmost block is for BHN)
the output variable- summing temperature. The central
block contains the rule base. Figure 7 shows the membership functions for input 1
i.e. required BHN.

Fig. 7: Membership functions for Brinell Hardness Number (input 1).

Figure 7 shows that three membership functions- 2) Membership Functions for Input 2 (% C)
softest, softer and soft are used for input 1. These
correspond to the hardness which a particular steel Figure 8 shows the membership functions for input 2
grade is required to acquire after the completion of i.e. %C.
annealing.

4
Fig. 8: Membership functions for percentage carbon (input 2).

Figure 8 shows that there are three membership 3) Membership Functions for the Output
functions- low, medium and high for input 2. These (sumtemp)
correspond to various types of plain carbon steel based
upon their chemical composition (carbon percentage). The membership functions of the output are shown in
figure 9.

Fig. 9: Membership functions for summing temperature (output).

Figure 9 shows that there are four membership process suitable for the present case is spheroidizing
functions- process anneal, spheroidize, full anneal and (refer figure 3).
normalize for the output. These correspond to the four The performance of the expert system was evaluated by
types of annealing processes which are to be performed testing the results obtained by its SIMULINK model
for the given steel. The output is added to the with the standard results for given set of inputs. Table
austenitizing temperature obtained from figure 2. The 3 shows the comparison of results obtained by the
final output is thus, the desired furnace temperature, Td. expert system with the standard values. It can also be
observed that full annealing of plain carbon steel having
IV. RESULTS & D ISCUSSIONS a carbon percentage of 0.6 requires the standard heating
Suppose that there is a requirement of heat treating a temperature of approximately 780ºC in order to acquire
steel having 0.6% carbon and after heat treatment the a hardness value of 210 BHN. The expert system gives
steel should acquire a BHN of 120. Now for such a case the desired temperature as 778ºC approximately (refer
we find that, the steel is a medium carbon steel (refer table 3). There is a difference of only 2ºC. Similarly, we
figure 8) and required hardness is softest (refer figure can find from table 3 that the values of Td set by the
7). After providing these two inputs to the expert expert system for different input conditions closely
system, it gives the austenitizing temperature as 700 match with the standard values of heating temperature.
deg.C. The expert system does it in the following Thus, we can say that, the designed expert system
manner. For the required BHN of 120 (softest) and %C works well for annealing of plain carbon steels.
as 0.6 (medium), the sumtemp (refer figure 9) is
approximately -50 deg. C. The sumtemp value proposes
that the given steel has to be spheroidized. From figure
2 it can be found that the austenitizing temperature
corresponding to plain carbon steel having 0.6 %
carbon is around 750 deg. C. The desired temperature
would be Td = -50 + 750 = 700 deg. C. The annealing

5
 Table 3: Table showing the comparison of results obtained by the expert system with the standard values

Heat treatment type Inputs Heated up to (t ºC)

Percent Brinell Standard As specified

Carbon Hardness value [3], by Expert
Number [4] System (Td)

Spheroidizing 0.6 120 700.12

0.9 120 700 750.2
Process Annealing 0.2 120 500 to 696.2
0.2 210 650/ 700 700.1
Full Annealing 0.6 210 780 777.6
0.9 210 830 827.7
0.2 350 860 855.4
Normalizing 0.6 350 805 808
0.9 350 855 858.1

IV. CONCLUSIONS
In this paper an attempt has been made to design an [5] D.A. Linkens, Minyou Chen, “Expert Control Systems- 2. Design
expert system that can predict the austenitizing Principles and Methods”, Engng Applic. Artif. Intell. Vol. 8, No. 5,
temperature or heating temperature of plain carbon 1995, pp. 527-537
[6] Donald R. Askeland and Pradeep P. Phule, “Essentials of
steels for their heat treament, especially annealing. The
Materials Science and Engineering”, Cengage Learning India Private
expert system is based upon Mamdani type fuzzy Limited, 2004, pp. 378-395
inference system that uses 6 set of rules. The results [7] www.efunda.com/process/heat_treat/softening/annealing.cfm,
obtained from the expert system were compared with January 13, 2012
[8] “Design Data”, PSG College of Technology, Coimbatore,
the standard results provided in the design data book for
Publisher- Kalaikathir Achchagam, 2007, pp. 2.1-2.3
the given set of inputs. It was found that the expert [9] Shu- Hsien Liao, “Expert system methodologies and applications-
system decided the correct values of temperature set a decade review from 1995 to 2004”, Expert Systems with
point for different types of annealing processes for plain Applications 28 (2005), pp. 93- 103
[10] Timothy J. Ross, “Fuzzy Logic with Engineering Applications”,
carbon steels. The proposed modeling approach can
John Wiley & Sons, Second Edition, 2010, pp. 178- 208
further be improved to predict the mechanical [11] Qian Zhang, Mahdi Mahfouf, “A hierarchical Mamdani- type
properties (tensile strength, ductility, hardness, etc.) by fuzzy modeling approach with new training data selection and multi-
correlating them with the heat treatment process objective optimization mechanisms: A special application for the
prediction of mechanical properties of alloy steels”, Applied Soft
conditions as well as chemical compositions using
Computing 11(2011), pp. 2419- 2443
complex, high-dimensional industrial data [11].

ACKNOWLEDGEMENT
We are grateful to Mr. N. Kumaraswamy (JRF, ITM
University) and Mr. L.N. Joshi (Technical Assistant,
PG and Mechatronics Lab., ITM University) in helping
us complete this work.

REFERENCES
[1] Natraj Mishra, M.Tech. Dissertation- “Design of an Expert
System for an Electric Muffle Furnace”, ITM University, Gurgaon,
Haryana, India
[2] GU Qiang, ZHONG Rui, JU Dong-ying, “Development of
materials database systemfor cae system of heat treatment based on
data mining technology”, Trans. Nonferrous Met. Soc. China
16(2006), s572- s576
[3] J.-S. R. Jang, C.-T. Sun, E. Mizutani, “Neuro-Fuzzy and Soft
Computing- A Computational Approach to Learning and Machine
Intelligence”, Pearson Education, Inc., 2008, pp. 27- 35
[4] D.A. Linkens, Minyou Chen, “Expert Control Systems- 1.
Concepts, Characteristics and Issues”, Engng Applic. Artif. Intell.
Vol. 8, No. 4, 1995, pp. 413-421