Neural network based control algorithm in robotised unmanned flexible manufacturing system.
Zuperl, Uros ; Cus, Franc ; Balic, Joze 等
1. INTRODUCTION
Detection of cutting tool condition is essential for faultless machining in flexible manufacturing systems (FMS). Unmanned Flexible
Manufacturing System (UFMS) is the most developed type of FMS. Such
system replaces human operator with robots, thus reducing labour costs
and prevents human errors. Because there is no human operator present,
the decision making system must monitor and control the whole process.
Some decision making system are commercially available, other are still
in the exploratory stages. Systems developed in laboratories, are often
multisensor systems embodying complex artificial intelligence
strategies. In commercially available systems, the one sensor approach
dominates. The main task of decision-making system is to analyze data
from sensors and to make appropriate control actions. The tool must be
in good condition during whole metal cutting process, therefore,
automatic detection of tool breakage is essential. Several different
approaches have been proposed to automate the cutting tool monitoring
process. These include classical statistical approaches (Bhattacharyya
et al., 2007) as well as fuzzy systems (Kuo, 2003) and neural networks
(Mulc et al., 2004). (Cus & zuperl, 2007) have used fuzzy expert
systems (Iqbal et al., 2007), fuzzy pattern recognition (Haber &
Alique, 2003), and fuzzy set theory for detecting tool wear and tool
breakage (Fu & Hope, 2006). The computer numerical control (CNC)
machine tools are not capable of tool breakage detection. They cannot
stop the process if the tool becomes damaged; therefore a monitoring
algorithm for unexpected tool breakage is developed. Neural networks
have been used often to monitor the progress of tool wear during milling
(Chien &Tsai, 2003) to predict the breakage of brittle tools, to
select the optimum cutting conditions and tool exchange cycle, to
estimate tool life by using flank wear measurements (Achiche et. al.,
2004). This paper presents a new monitoring system using neural networks
to monitor tool failure in real time. The neural decision making system
is used with cutting force and machining parameters as input factors.
The simulation and experimental results showed that this system is
accurate enough for monitoring abnormal tool states in real time.
[FIGURE 1 OMITTED]
2. MONITORING AND CONTROL STRUCTURE
The proposed monitoring system consists of sensor, signal
processing module, data acquisition module, monitoring module and
decision making systems to interpret the sensory information and to
decide on the essential corrective action.
A 3 component piezoelectric dynamometer (Kistler 9255) is selected
to monitor the cutting forces in the X and Y directions. An analogue
signal from the force sensor is converted to a digital form. Signal
generated by the sensor and conditioned by amplifier is send through
data acquisition module to the signal processing module and monitoring
module, which is directly connected to the machine control (CNC).
It provides reliable detection of tool and process failures. A
signal processing module has the following properties: very accurate
discrimination of state condition, easy application to pattern
classifier, and high speed of signal processing.
AR model and band energy using digital filters were selected as
signal processing technologies. Features extracted from signal
processing are used as inputs to a pattern classifier, which decides the
state conditions. The signals were monitored using a fast data
acquisition card (National Instruments PC-MIO-16E-4) and software
written with The National Instruments CVI programming package.
The system development course consists of two main steps: First, a
neural network tool wear model is developed from a set of data obtained
during actual machining tests performed on a Heller milling machine
using a Kistler force sensor. The relationship between the machining
parameters/sensor signals and flank wear is captured via this network. A
well known neural network having a back-propagation learning algorithm
is then used as a decision making system to monitor tool failure. Before
the use of neural network it was necessary to teach the network the tool
states.
Figure 1 shows the basic architecture of the proposed system. A
neural decision making module was developed in Matlab software.
Network has two hidden layers and uses a set of 5 normalized inputs
for tool condition prediction: (1) cutting speed, (2) feed rate, (3)
depths of cut, (4) forces, (5) tool wear. Output layer consist of only
two neurons: (1) normal and (2) broken/worn. It has tool-breakage
detection capability and is based on pattern recognition. The neural
network stores a number of reference force patterns that are
characteristic of tool breakage. When a tool tooth breaks, pattern is
identified and a break is declared within 10 ms of the breakage. The
goal of developing the control algorithm of TCM is not only to produce a
reliable, but also as cheap as possible monitoring system.
3. EXPERIMENTS, RESULTS AND DISCUSSION
The monitoring experiments involve an end milling process of steel
parts using two end mill cutters: normal and on tooth broken. The
vertical machining center Heller BEA 02 used has a maximum spindle speed
of 15.000 rpm, a minimum feed resolution of 5 urn and feed rate of 35
m/min.
The force sensor is filtered by the band-pass filter from 0.5 to
1.5kHz.The experiment was conducted using a four edge milling cutter
with 12 mm diameter. The workpiece material used in the machining test
was Ck 45 and Ck 45 (XM) with improved machining properties. The
workpiece material used in the machining test was Ck 45 and Ck 45 (XM)
with improved machining properties. The neural network was capable of
detecting tool conditions accurately in real time.
[FIGURE 2 OMITTED]
The accuracy of training data was 97.8%, and the accuracy of
testing data was 91.2%.
The output node value of a back-propagation neural network was
mapped as 0.01 for the normal cutting state, and 0.99 for the tool
breakage.
When the neural network outputs are over 0.9 (tool breakage), it
sends the signal "Tool broken" to the display. When both the
neural network outputs are below 0.9, it sends the signal "Tool
condition Normal".
Developed decision system incorporates dynamics limits for tool
breakage detection. The two dynamic limits above and below the monitor
signal follow the monitor signal continuously (Figure 2).
Slow but large load changes due to variations in cutting depth
(hardness, oversize) are tolerated at a ratio up to 1:4.
4. CONCLUSION
In robotised flexible manufacturing system a monitoring system is
developed that can detect/control tool breakage and chipping in real
time by using a combination of artificial neural networks.
An intelligent monitoring module is used to extract the features of
tool states from cutting force signals.
The proposed control algorithm of TCM is easy to install in
existing or new machines, and do not influence machine integrity and
stiffness. The simulation and experimental results showed that this
system is accurate enough for monitoring abnormal tool states in real
time.
Future research should consider applying different intelligent
decision making techniques, such as fuzzy logic, genetic algorithms,
genetic programming and ANFIS to see which technique is the most
accurate and reliable.
5. REFERENCES
Achiche, S.; Balazinski, M.; Baron, L. & Jemielniak, K. (2004).
Tool wear monitoring using genetically-generated fuzzy knowledge bases.
Engineering Applications of Artificial Intelligence, Vol. 15, 303-314
Bhattacharyya, P.; Sengupta, D.; & Mukhopadhyay, S. (2007).
Cutting force-based real-time estimation of tool wear in face milling
using a combination of signal processing techniques. Mechanical Systems
and Signal Processing, Vol. 21, 2665-2683
Chien, W. T. & Tsai, C. S. (2003). The investigation on the
prediction of tool wear and the determination of optimum cutting
conditions in machining 17-4PH stainless steel. Journal of Materials
Processing Technology, Vol. 140, 340-345
Cus, F. & Zuperl, U. (2007). Adaptive self-learning controller
design for feedrate maximization of machining process. Adv. Produc.
Engineer. Manag., Vol. 2, No. 1, 18-27
Fu, P. & Hope, A. D. (2006). Intelligent Classification of
Cutting Tool Wear States. Advances in Neural Networks, Vol. 39,
3211-3349
Haber, R. E. & Alique, A. (2003). Intelligent process
supervision for predicting tool wear in machining processes.
Mechatronics, Vol. 13, 825-849
Iqbal, A.; He, N.; Dar, N. U. & Li, L. (2007). Comparison of
fuzzy expert system based strategies of offline and online estimation of
flank wear in hard milling process. Expert Systems with Applications,
Vol. 33, 61-66
Kuo, R. J. (2003). Multi-sensor integration for on-line tool wear
estimation through artificial neural networks and fuzzy neural network.
Engineering Applications of Artificial Intelligence, Vol. 3, 49-261
Mulc, T.; Udiljak, T.; Cus, F. & Milfelner, M. (2004).
Monitoring cutting- tool wear using signals from the control system.
Stroj. vestn., Vol. 50, No. 12, 568-579
