Você está na página 1de 6

1.

Introduction
LabVIEW is a powerful graphical programming language developed by National
Instrument that has its roots in automation and data acquisition. It is becoming very
popular lately. While for neural network, it is considered as a very powerful method
for process identification and control of nonlinear system. Deploying both of them is
considered to be necessary as a further toolset for researches.
Currently, National Instrument doesn’t provide toolbox for neural network
controller. Using third party toolbox will require extra spending. There are many papers
and tutorials discussing about neural network implementation in LabVIEW but they will
come with a very limited or very simple source code.
This work basically is a reimplementation of a conference paper: “LabVIEW based
Intelligent Control for Speed Regulation of Electric Motor” by Dr. Jerome et al. For more
detail information, you can refer directly to this paper.

2. Design
In this work, a typical 2-layer feed forward neural network is used. The activation
function of the input layer is an identity function and hidden and output layer has a
sigmoid function.

Figure 1 Neural Network Architecture

2.1.Feed Forward
The network will receive single input and send single output. The input signal will be
broadcasted to each neuron in the next layer.

Direction of signal
2.2.Back Propagation
The next step is back propagation of errors. During this step, the weight is adjusted
to minimize the errors. ∑
δ(1)()
dηδ
+=+
=−
=−−iiiiδ
wtww
wtwtv
vvw
yyyy
(1)
i()
jjjiij
jkjktjij
jijij

3. Implementation
3.1.Neural Network Block Diagram
The neural network algorithm is implemented into a subVI block. Here is the
block diagram. This block diagram is implemented based on the previous equations.

Figure 2 Neural Network Block Diagram

3.2.DC Motor Model


In this work, the neural network controller is used to improve DC drive
performance. It is expected that this neural network controller can achieve accurate
trajectory control of speed when DC drive and load dynamics are unknown. The DC
motor is driven with PWM (Pulse width Modulation) method where the speed can be
controlled by using the duty cycle of the PWM.
Figure 3 DC Motor Model

.
DC motor parameters used for this work are: Ke =3.475Nm/A, J = 0.068 kgm2, Ra = 7.56
ohm and La = 0.055 H. Hence, the transfer function of the DC motor can be stated as
follow.

3.3.Structure of ANN Controller for DC Motor


For the controller, two neural network systems are deployed. One will receive the
desired speed as input and another one will receive error as input. Both will send PWM
duty cycle as the outputs. These values will be summed and the result will be the control
signal for the DC motor.

Figure 4 Neural Network Controller Architecture

4. Simulation and Results


4.1.Simulation
The next step is to train the neural network. Through this training, the plant
dynamics will be identified by the neural network. The objective is to achieve identical
behavior between the plant and the neural network. So, PWM Duty Cycle will be given as
the input and the actual motor speed as the desired output.
Input: Desired Output
:
PWM Duty Cycle Actual Motor Speed

Figure 5 Training Process

The training process will also be conducted in LabVIEW. While on training it is


necessary to adjust the bias to reduce the error that occurs. This is done by trial and error
by first setting the bias close to zero. When the neural network has already converged to a
certain value while it is still showing an intolerable error, then this bias value can be
increased gradually.
Here is the LabVIEW block diagram for training process.

Figure 6 Block Diagram for Training

Next, the trained neural network is applied into the system. Here is the block
diagram.

Figure 7 Block Diagram for Normal Running


4.2.Results
After training, the best results are achieved with these parameters:
• Weight input 1 = -3.12317
• Weight Input 2 = -7.0356
• Weight Bias Input 1 = -0.211902
• Weight Bias Input 2 = -0.211902
• Weight Output 1 = -3.12317
• Weight Output 2 = -7.0356
• Weight Bias Output = 11.8467
• Input Bias = 0
• Output Bias = 0.2
With those values, the neural network controller shows a good performance in term of
input tracking capability. Here are the results:

5. Conclusions
From this final project, the conclusions that can be made are as follows.
• Neural network can be implemented on LABVIEW.
• Simulation result shows that neural network controller can give a very good
performance.
6. References
• Jerome, et al, LabVIEW based Intelligent Controllers for Speed Regulation
of Electric Motor, Instrumentation and Measurement Technology Conference
Ottawa, Canada, 17-19 May 2005.
• Prasanth Kumar, Intelligent Control Lecture Note, School of Mechanical and
Aerospace Engineering, Gyeongsang National University, Republic of Korea.
• [ONLINE]. Available: www.generation5.org

Você também pode gostar