The instrument should also be designed to be able to calculate the concentration of additives in the liquid. For this purpose, a multi-sensor model including capacitive liquid level sensor (CLS), ultrasonic liquid level sensor (ULS) and capacitive pressure sensor is used to obtain liquid information. The data acquired from all these sensors is processed using Pau's multi-sensor data fusion framework to calculate the level of the liquid and the concentration of additives added to the solution. Pau's framework includes alignment, correlation functions, analysis and presentation functions. The designed multi-sensor technology has been tested with real-world data for a variety of fluid levels and additives. The obtained results show that successful implementation of the proposed objectives is possible with a root mean square percentage error of 1.1% over the entire range.

1. Start

The flattening process is one of the most common processes in any process industry such as diary, pharmaceutical, fabric, petroleum and confectionery. Liquid processes ideally involve controlling the level of liquid in a vessel at a user-desired set point level under dynamic or static conditions. The controller is implemented to obtain the desired set point by actuating final control elements. In order to design the controller, an instrument is required to measure the actual liquid level. The instrument is a device used to measure a physical parameter (in this case, liquid level).

Many measurement technologies are available for liquid level measurement.A detailed study of available measurement techniques is reported.

2. Experimental device

The experimental setup consists of a tank that stores liquid.To measure the level of a liquid, acapacitive level sensor.The output of this sensor is connected to a signal conditioning circuit to convert the output signal into a standard voltage signal.Then use the NI myRIO board to obtain the voltage signal, as shown in the figure1Shown.

Figure 1 

Experimental setup.

The equation for a capacitive level sensor is derived from the basic principles of capacitance, where the capacitance between two parallel plates can be given in the equation (1)[ 40, 41 ].Among them " "is the cross-sectional area of the plate," " is the distance between the two plates, " " is the dielectric constant of air, and " " is the dielectric constant of liquid. If the dielectric component changes, the capacitance will change accordingly.

In the proposed work, the design of the capacitive level sensor considers two parallel cylindrical rods acting as electrodes instead of parallel plates.As shown in the picture2As shown, this structure was chosen for ease of maintenance and increased durability.

Figure 2 

Schematic diagram of a capacitive level sensor.

Total capacitanceBy[ 42 ] givesAmong them,Is the total capacitance, is the vacuum dielectric constant, is the relative dielectric constant of water, is the distance between the electrodes, is the electrode radius, is the liquid level, and is the height of the electrode and container.

In the experimental setup considered for the proposed work, the dimensions of the capacitive level sensor are the distance between the electrodes , Electrode radius, And the full height of the tank.

To convert the CLS output into voltage, a timer circuit and frequency to voltage conversion circuit are used.The timer circuit includes a3555 IC shown,And a frequency to voltage conversion using LM331 IC and a picture as shown4The conversion circuit shown.The outputs obtained from the timer circuit and frequency to voltage circuit respectively are given by the equations (3) and the equation (4) means.The output voltage is then collected to a computer using the analog channels of an NI myRIO (National Instruments reconfigurable input and output embedded board).The data sheets of the 555 IC and LM331 IC provide the following equations.Figure5Figure 4 shows the device for acquiring data via [ 43 ]Get the capacitance and convert to voltage to get the actual implementation of the circuit to PC.

Figure 3 

Timer circuit diagram.

Figure 4 

Circuit diagram of frequency to voltage converter.

Figure 5 

Practical settings for signal conditioning and acquisition.

3.Problem statement

The previous section shows how the experimental setup is designed.The experimental setup was placed at different liquid levels and the output power was plotted.To conduct the experiment, we used water and NI myRIO collected output at different levels, which is shown in the figure.6Shown.As shown in the picture6Shown, the highlighted area from the characteristic diagram of the capacitive level sensorIt can be seen that part of the initial measurement range has higher nonlinearity.This effect in the measurement is related to the mechanical design of the electrode, as shown in the figure of the electrode used7Shown.

Figure 6 

The output of a capacitive liquid level sensor, used to change liquid level.

Figure 7 

Capacitive electrodes for liquid level measurement.

To understand the effect of solution concentration on capacitive sensor output, sugar solution was used instead of water.The same test results are shown in the table1Shown, the complete changes in capacitance, frequency and voltage are shown in the table2Shown.A similar analysis was performed on changes in liquid concentration by adding 10%, 20% and 30% sugar solutions.Figure8Shows the CLS output as a function of concentration and level.From table1Obviously, the output of the capacitor is a function of the liquid level and the properties of the liquid (dielectric constant).This characteristic of capacitive level sensors is used in the proposed work to design a multi-sensor system that can measure the level of a liquid and additionally indicate detailed information about the concentration of the liquid under consideration.

Table 1 

The dielectric constant changes with changes in sugar concentration.

Table 2 

The output of a capacitive level sensor is used to change the concentration of a solution.

Figure 8 

Plot of CLS output changes as sugar solution concentration changes.

4.Problem solving

In order to achieve the purpose of measuring liquid level and concentration simultaneously, in addition to using the availableCapacitive level sensorIn addition, a multi-sensor system is integrated.As shown in the picture9As shown, two additional sensors (ultrasonic level sensor and pressure sensor) are placed on the container where the level is to be measured.We consider a transport-type ultrasonic level sensor that measures the level of a liquid based on the time it takes for the signal to travel back from the sensor and back.This value will directly indicate the fluid level and will not vary with fluid type.In addition, a pressure sensor is used in the experiment, which will measure the liquid level based on the density of the liquid.

Figure 9 

Process setup with multiple physical sensors.

Data acquired from these three sensors were combined using Pau's framework [ 44 ].Figure10'sBlock diagramShows the processing flow of data fusion using Pau.The first step in the data fusion process is data alignment.At this stage, it can be seen that the data to be fused will be of the same size and dimensionless.

Figure 10 

Pau's framework.

4.1. additional sensors

(1)ultrasonic level sensor. Transit Time Ultrasonic Level Sensors (ULS) are used to measure the level of liquids. Ultrasonic level sensors are placed on top of the vessel so that waves are transmitted over the surface of the water and the reflected signal is captured back. The time it takes for the signal to travel and reflect is used to calculate the level of the liquid. The timer port of the microcontroller is used to calculate the travel time and the output corresponding to this time is transmitted in the form of a 16-bit digital signal. Standalone NI myRIO acquires 16-bit data related to fluid level(2)capacitive pressure sensor. The output of a capacitive pressure sensor (CPS) changes as the input pressure changes. As the liquid height changes, the pressure developed at the nozzle changes. When the liquid level is high, the liquid pressure is greater, and as the liquid level in the tank decreases, the pressure decreases accordingly. This change in pressure also changes with the change in density of the liquid. table3Density changes with changing sugar concentration are shown. The experimental setup consists of a.

From table

3

It is obvious that changes in sugar concentration cause changes in liquid density. Changes in pressure cause changes in the physical dimensions of the diaphragms used in CPS, causing relative changes in capacitance. This change in capacitance is converted into voltage using a De Sauty bridge and rectifier circuit. The voltage output by the rectifier is obtained from the analog channel of NI myRIO, as shown in the figure

11

Shown. From the picture

11

As can be seen in, there is nonlinearity in the characteristics due to the accuracy of the discrete components, and the use of artificial neural networks during fusion can compensate for the nonlinearity.

Table 3 

Change the concentration to change the liquid concentration.

Figure 11 

The output of the capacitive pressure sensor is a signal conversion output.

4.2. Alignment technology

Now that the data from all three sensors (CLS, ULS, and pressure sensor) have been acquired onto the NI myRIO board, they need to be fused and processed.The first step in convergence is to align technologies/common representation formats.The purpose of this phase is to ensure that all sensor sources have equal priority in terms of size.This means that all signals will be directed to a dimensionless value ranging from 0 to 1 or 0 to 100%.In the proposed work, we consider it to be between 0 and 1. Since the data obtained from all three sources are one-dimensional, the radiometric normalization technique is implemented.(1)Alignment technology for CLS output. The output produced by a capacitive level sensor with a data conversion circuit is a change in voltage as the level changes. The voltage changes from 0.05 V to 0.35 V and the liquid level changes from 0 cm to 50 cm. A spatial transformation is used in the proposed work to convert the signal intensity from 0 to 1. For spatial transformation, the signal level of 0.05 V is taken as a reference and the rest of the signal is calculated using this point(2)Alignment technology for ULS output. Timer data corresponding to the transmission time of the ultrasonic signal representing the liquid level is converted into a range of 0 to 1. To do this, we use a spatial alignment technique, where the time taken for wave transmission and reflection is considered "0" when the tank is full and "1" when the tank is empty. Extrapolate the intermediate value based on the time it takes for the signal to travel and reflect(3)Alignment Techniques for Pressure Sensor Outputs. From the picture11The input-output relationship can be seen that the output of the pressure sensor is a voltage change from 0.1 V to 3.6 V. In order to represent this data with a normalized value between 0 and 1 (as shown in the figure12Shown), using the equation (5) represents the min-max expression.

Figure 12 

Capacitive pressure sensor CRF output.

4.3. association integration

Correlation is a technique in which data is grouped into distinct clusters across the input range to perform calculations related to a target. Group clusters so that the contribution of that particular cluster of data to achieving the target functionality can be calculated. For this purpose, some features are used such as nearest neighbor, boundary conditions, binary trees and kd-trees. In the proposed work, boundary condition algorithm is used for CLS data; kd-tree algorithm is used for pressure sensor data, followed by nearest neighbor algorithm for ULS data.

4.4. analysis and aggregation or decision-making process

The decision-making step in the proposed work involves the calculation of the liquid level, independent of the liquid used to determine the solution concentration.In the presented work, evidence matching technique is used.For evidence matching, a neural network algorithm is used [ 45 ].For training the neural network, data corresponding to liquid levels with different defined solvent concentrations are considered.Here, add 50g, 100g, 150g and 200g of sugar per 10 liters of water.

The input data to the neural network module comes from correlation vectors, each from a capacitive level sensor, a pressure sensor and an ultrasonic level sensor.The neural network algorithm is trained so that information related to the liquid level is derived directly from the output of the ultrasonic liquid level sensor.For information corresponding to added liquid, CLS and pressure sensor data are used.Since it is clearly shown that the CLS output depends on the level of the liquid and the type of liquid (dielectric constant of the liquid), the output of the pressure sensor depends on the density (volume and mass) of the liquid.The proposed work adopts the backpropagation network of artificial bee colony algorithm.The backpropagation network can be decomposed into feedforward calculation and backpropagation to the output layer.Propagate backward to the hidden layer and update the weights.TableFigure 4Details about the neural network model implemented in the proposed work are shown.A diagram of neural network signal connectivity is shown in the figure.13Shown.

Table 4 

Neural network parameters.

Figure 13 

Neural network connectivity and structure.

The weight of the first connection matrix is ,  .Likewise, the weights of the second connection matrix are,  , There will be ()The weight between the input side and the hidden layer, similarly, ()The weights are between the hidden layer and the output.By considering the associated vectors of the three transducer signals as, InspireForThe hidden unit is given by 

In an S shape" "As an activation function, a forward vector is calculated.

In the reverse path, starting from the output of the network, the computation is similarly reversed, minimizing the error.Artificial bee colony algorithm is used to update weights.Training continues until the network produces targets by target.This is through monitoringThis is achieved by a regression value " " which should ideally be 1 and a mean square error " MSE" which should ideally be zero.

4.5. express

At this stage, the data is converted into the form of physical variables. Before this stage, calculations are performed using dimensionless values in the range 0 to 1. The results obtained from the neural network will also be in the range of 0 to 1, which may be difficult for users to understand. Therefore, the results must be expressed in terms of liquid level height in "cm" and liquid concentration in "grams per liter".

5. Results and discussion

Once the data fusion process is complete, the performance of the reported techniques needs to be evaluated.To test, measure the level of the liquid by varying the concentration of sugar used in the water and changing the level of the liquid.The level of the liquid varies from 0 to 50 cm, and subsequently, the sugar concentration changes from 0 g/L of water to 25 g/L of water.Table5Shows the same results.

Table 5 

Results of water level measurements at different sugar concentrations.

Level measurement of liquids with varying concentrations using designed multi-sensor technology.Measure liquid levels for different liquid concentrations.On the table5 medium, recording the output of the instrument for varying sugar concentration measurements.When different sugar concentrations are used and different liquids are used, the liquid level can be measured without recalibrating the system.In any instrument available, errors are introduced when changing process media, this is not the case here.The proposed system is used to measure different liquid levels with different liquid concentrations from 0 m to 50 cm.From table5It can be seen that the proposed technique is able to measure the liquid level with a maximum deviation of 5 mm from the expected value.The proposed system is also designed to identify changes in concentration as a means of measuring additives in water.Table5The last column ofThe measured sugar concentration in the solution is shown.It can be seen that the proposed technique can measure the concentration of sugar with a maximum deviation of 400 mg from the actual value.

The measured characteristics expressed as percentage error were obtained and plotted in Fig.14 and 15 medium,For liquid level measurement and concentration analysis.It can be seen that the maximum percentage error produced by the measurement system is approximately 7%.

Figure 14 

Error characteristics of syrup level measurement.

Figure 15 

Error characteristics of syrup concentration analysis.

To evaluate performance, measurements were made taking into account milk powder instead of sugar.Table6Shows the results obtained from the tests carried out.For the measurement of liquid level, the range considered is 0 cm to 50 cm, and the concentration of milk powder per liter of water varies between 0 g and 100 g.The error characteristics in Fig.16 and 17Medium draw.

Table 6 

Water level measurement results for different milk powder concentrations.

Figure 16 

Error characteristics of milk level measurement.

Figure 17 

Error characteristics of milk powder concentration analysis.

Similar to the experiments using sugar solutions, the proposed system was tested by considering milk solutions to verify the proposedMeasurement technology. Similar to the previous experiment, the results showed a maximum deviation of 1 cm in the measured liquid level and 3 gm/L in the measured concentration compared to the actual value.

From table5 and 6And pictures14 - 17, it can be seen that the reported multi-sensor technology is able to accurately and robustly measure liquid level without the influence of additives in liquids.Additive concentrations can also be accurately measured.Except table5And table6In addition to the results shown, further tests were carried out on approximately 260 samples under different conditions (different concentrations of additives and liquid levels).The percentage root mean square error derived from the reported measurement technique is 1.1%.