1.3 CONTEST CATEGORY:
Student
2. BENEFITS OF USING VIRTUAL INSTRUMENTATION:
A. Instrument can be custom made with desired features for specific application
B. Errors can be easily identified and rectified, while for real instruments repairing and
calibration needs to be done regularly.
C. Easy to use, can be put to use by students, engineers, as well as scientists.
D. Reduced cost of running in the long run.
E. Reduced application development time.
3. NI HARDWARE AND SOFTWARE USED IN THIS APPLICATION:
The choice made for the data acquisition and processing software was NI LabVIEW 8.5.1
and the NI hardware used was NI-DAQ 6024E.
4. PROBLEM TO BE SOLVED:
To automate a reverse osmosis plant so as to increase its efficiency and maintain some
output parameters like product water flux, salt-rejection ratio, product recovery rate at
desired levels.
5. SOLUTION TO THE PROBLEM:
Using the NI-DAQ 6024E hardware and LabVIEW 8.5.1 the plant was automated to
meet the specific requirements on output water quality.
6.INTRODUCTION TO APPLICATION:
The reverse osmosis plant has basic units called membrane modules shown above and
also the real module in Fig.4. Water is pumped into the module using a high pressure
pump and desalinated water is collected at the output.
The reverse osmosis plant for which the system is developed was being operated
manually. For instance, suppose the product water flux decreased, then a supervisor
would increase the pressure till the output flowmeter gave the desired reading.This
method of control was found to have a degrading effect on the plant output and efficiency
over time. This was because the reverse osmosis process is a very complex chemical
process with multiple input parameters like the feedwater flowrate, feed pressure, feed
temperature, feedwater salinity, feedwater pH which are constantly fluctuating
throughout the day and year, which in turn gives rise to controlled variable fluctuations
like unsteady product water flux, and product salinity.
Such a situation was found ideal to implement automatic control in order to increase the
efficiency and maintain throughput of the plant as well as keep track of healthiness of the
plant in terms of membrane fouling due to salt deposition and compation due to constant
subjection to high pressures.
7.APPLICATION DESCRIPTION:
The choice made for the data acquisition and processing software was NI LabVIEW 8.5.1
and the hardware used was NI-DAQ 6024E. A model referenced control system was the
choice to be implemented. To make a control system a model of the plant was to be
developed (Fig.1). Modeling the plant to develop a model referenced control system was
a Multiple Input Multiple Output (MIMO) function approximation task i.e we had to find
that function f in the equation ,
Output parameter = f (input parameters)
Mathematical modeling was tried out using the chemical equations developed in
chemistry literature for reverse osmosis process which proved to be inaccurate due to the
complexity and various disturbances present in the chemical process of reverse osmosis
which the equations neglected.
A VI was then developed to model the plant using the radial basis function neural
networks which are best suited for function approximation problems. The training data
was acquired by sampling the numerous input parameters such as feed water temperature,
feed pressure, feed salinity, feed flowrate, feedwater pH by varying them over
permissible ranges and also the corresponding output parameters like product water flux,
product salinity.
The neural model reference control architecture uses two neural networks: a controller
network and a plant model network, as shown in the following figure. The plant model is
identified first, and then the controller is trained so that the plant output follows the
reference model output.
The model error shown in figure was used to train the plant model network. Having
trained the plant model neural network the plant model was ready. After this the
controller was trained using the dynamic backpropagation algorithm using the control
error. The dynamic backpropagation algorithm backpropagates the control error through
the plant model without readjusting any weights. This backpropagated error is then used
to train the controller network i.e to adjust its weights and biases.
The plant healthiness was tracked using the model error. This was because the plant
model developed at the beginning of operation modeled the plant without any membrane
fouling or compaction. Over time as fouling and compaction degrades membrane
performance the model error automatically will rise thus giving us a measure of the
deterioration in plant health.
Fig.1.Neural model reference control architecture
The training algorithm for both the plant model and controller was programmed using
LabVIEW 8.5.1 using the LabVIEW Mathscript structure.
8. SYSTEM SETUP OF THE APPLCATION:
Fig. 2. The Single Stage RO Plant To Be Controlled.
Fig. 3. Computer Interfacing Through NI-DAQ 6024E
Fig. 4. The RO module used in the single stage RO plant.
Fig. 5. The computer control system.
Fig. 6. Various sensors on input/output flow lines.
The system setup includes the neural control system implemented in NI LabVIEW 8.5.1
interfaced to the plant through the NI-DAQ 6024E. Then there is a power supply unit for
the sensors. The sensors perform the sensing on the input and output flowlines of the
module and collect the input and output parameter data. Those parameters are :
A. Input parameters : Feedwater flowrate, Feedwater pH, Feedwater pressure, Feedwater
temperature, Feedwater salinity.
B. Output parameters: Product flowrate, Product salinity.
The manipulated variable is the pressure generated by the high pressure pump.This
control action i.e pressure is set by the neural controller through the NI-DAQ 6024E
analog channels.
9.SOFTWARE IMPLEMENTATION IN THE APPLICATION:
Fig. 7.Front panel of the SubVI which models the plant
Fig. 8. Block Diagram Of The SubVI which models the plant.
Fig. 9. Block Diagram Of SubVI which generates a neural network.
Fig. 10.User Defined M-files Used By Mathscript Node.
The algorithm for generating the neural network and training them was done using the
LabVIEW Mathscript structure in LabVIEW 8.5.1.
The VI developed iteratively creates a radial basis network one neuron at a time. Neurons
are added to the network until the sum-squared error falls beneath an error goal or a
maximum number of neurons has been reached. At each iteration the input parameter
vector that results in lowering the network error the most, is used to create a radial basis
neuron. The error of the new network is checked, and if low enough the VI is finished.
Otherwise the next neuron is added. This procedure is repeated until the error goal is met,
or the maximum number of neurons is reached. This VI whose panels are shown above
was used to generate both the plant model network as well as the controller network.
MS excel worksheets were used for datalogging purpose. Data was read from the
worksheet for processing and then the control action was sent out using the data
acquisition functions.
10.NATIONAL INSTRUMENTS PRODUCTS PART OF APPLICATION:
NI-DAQ 6024 played a major role in data acquisition and control while the LabVIEW
8.5.1 version helped implement the neural control algorithm with ease.
11.SPECIFIC CAPABILITIES OF NI PRODUCTS USED:
The LabVIEW Mathscript structure available in LabVIEW 8.5.1 was crucial in easy
implementation of the algorithm. The friendly user interface to control the plant was a
bonus. Also the datalogging facility helped in keeping history information about the plant
performance.
12. TIME, MONEY SAVINGS:
The number of staff working at the station was cut down to about 50% of original as a
result the salary cost was reduce to half. Also the regular inspections had to be done
during which the plant had to be shut down. However now the inspection is automatic,
example if the RO membrane has scaled beyond a certain limit due to salt deposition on
it, then backwashing of the membranes is automatically done whenever detected. Also
savings are made as the RO module life time has been found to have increased.
13.ACKNOWLEGEMENT:
The described project is an overview of the bigger 18 months project worth 18 lakhs
sponsored by Central Salt & Marine Chemicals Research Institute (CSMCRI),
Bhavnagar.
We would like to thank Dr.Chandrashekhar, Director CEERI and Dr.L.K.Maheshwari,
Director BITS-pilani for providing us with an opportunity to work together on this
project.
14.CONCLUSION
An appropriate automation for a reverse osmosis plant is developed using radial basis
function neural networks developed in LabVIEW 8.5.1 unlike the traditional approach
of using mathematical modeling of plants for control.
No comments:
Post a Comment