WO2007096849A1

WO2007096849A1 - A voltammetric analysis system

Info

Publication number: WO2007096849A1
Application number: PCT/IE2007/000021
Authority: WO
Inventors: Barry O'connor; Kilian Murphy; Axel Rau; Karen Twomey
Original assignee: University College Cork - National University Of Ireland, Cork
Priority date: 2006-02-20
Filing date: 2007-02-19
Publication date: 2007-08-30

Abstract

An analysis system (1) has a sensing head (2) having voltammetric working(12) and counter electrodes (10), a reference electrode (11), a signal drive and capture circuit (3), and a processor (4) for processing response data. The processor (4) pre-processes the response data to reduce its dimensionality, trains an artificial intelligence model, and applies the reduced data to the model to provide analysis results. The processor (4) pre-processes the response data by determining a shape descriptor to represent the response data, such as a value for voltammagram moment. The model is a neural network model.

Description

A VOLTAMMETRIC ANALYSIS SYSTEM

INTRODUCTION

Field of the Invention

The invention relates to analysis of fluids such as liquid foods or monitoring of inline cleaning processes.

Prior Art Discussion

It is known to use voltammetry to analyse fluids. For example, US6841053 describes an electronic tongue for detecting ozone. It comprises a working electrode and a counter electrode, a pulse generator for applying energizing pulses to the working electrode, a recoding device for recording output from the working electrode, a sampling device for sampling the output, and a processing unit for performing a multivariate analysis of a matrix of sampled values.

The invention is directed towards providing for improved identification of composition of a fluid.

SUMMARY OF THE INVENTION

According to the invention, there is provided an analysis system comprising: a sensing head having voltammetric working and counter electrodes, a reference electrode; a signal drive circuit for applying potentials across the working and reference electrodes when they are in contact with a sample;

•a capture circuit for capturing counter electrode current voltammetric response data, and a processor for processing said response data, wherein the processor comprises means for pre-processing the response data to reduce its dimensionality, for training an artificial intelligence model, and for applying the reduced data to the model to provide analysis results.

In one embodiment, the processor pre-processes the response data by determining a shape descriptor to represent the response data.

hi one embodiment, the shape descriptor includes a value for voltammagram moment.

hi one embodiment, the shape descriptor includes a value for shape voltammagram area.

In one embodiment, the shape descriptor includes a value for voltammagram orientation.

In one embodiment, the shape descriptor includes a value for voltammagram compactness.

In one embodiment, the shape descriptor includes a value for voltammagram eccentricity.

In one embodiment, the model is a neural network model.

hi one embodiment, the neural network model is of the multi-layer perception type.

hi one embodiment, the data processor comprises means for training the model by successively applying signals across the working and reference electrodes, measuring resulting currents in the counter electrode, generating data corresponding to said measured currents, and applying training data to a transformation matrix.

In another embodiment, the sensing head comprises the reference electrode co- mounted on a support also supporting the counter and working electrodes. In one embodiment, the counter electrode is centrally mounted, and the reference electrode is annular, surrounding the counter electrode.

In one embodiment, the sensing head comprises a plurality of working electrodes.

In one embodiment, the working electrodes surround the counter electrode and are between the counter electrode and the reference electrode.

In one embodiment, the counter electrode is disc-shaped having a centre aligned with a sensor head axis, the working electrodes are disc-shaped and are equi-distant from the axis, and the reference electrode is annular and planar and is symmetrical about the axis.

In another aspect, the invention provides a computer readable medium comprising software code for performing operations of the processor of any system defined above when executing on a digital processor.

DETAILED DESCRIPTION OF THE INVENTION

Brief Description of the Drawings

The invention will be more clearly understood from the following description of some embodiments thereof, given by way of example only with reference to the accompanying drawings in which:-

Fig. 1 is a high level block diagram of an analysis system of the invention;

Fig. 2 is a perspective view of a sensing head of the system;

Figs. 3 to 5 are plots of voltammetry input signals;

Figs 6 to 8 are flow diagrams illustrating operation of the system; and Fig. 9 is a plot illustrating compositions of 35 sample liquids analysed by the system;

Fig. 10 is a plot of a raw data voltammagram output;

Figs. 11 to 22 are example plots of reduced data;

Fig. 23 is a plot of result data for each of 35 samples; and

Figs. 24 and 25 are calibration plots.

Description of the Embodiments

Referring to Figs. 1 and 2, an analysis system 1 of the invention comprises a sensing head 2, an electronic controller 3, and a computer 4. The sensing head 2 comprises a central disc-shaped counter electrode (CE) 10, an annular and planar reference electrode (RE) 11, five planar disc-shaped working electrodes (WE) 12, and a Kynar™ cylindrical housing 13. The working and counter electrodes are embedded in epoxy which can withstand harsh operating conditions such as high temperature and pressure. Sample solutions contact the entire bottom face of the sensing head 2.

The reference electrode 11 of silver/silver chloride (Ag/AgCl material). The counter electrode 10 is made of stainless steel. The system 1 uses an array of noble metals as its working electrodes 12, in this embodiment Gold (Au), Iridium (Ir), Platinum (Pt) and Rhodium (Rh).

For each electrode a range of input parameters may be applied and incorporated in a single measurement.

There is a membrane covering the RE 11. The reference electrode 11 is inbuilt in the sensing head 2 instead of being a separate discrete item as typically in the prior art. This arrangement ensures that the reference electrode 11 is equidistant from each working electrode 12 and allows an accurate comparison of the currents that develop at each working electrode.

An operator uses the PC 4 to set the parameters of the signals to be sent to the sensing head 2. The PC 4 is also used to select the working electrodes 12 to be used. Up to five WEs 12 can be chosen for a single measurement. When instructed, the PC 4 sends the parameters for each electrode to the electronic controller 3. The controller 3 is programmed to generate the required signals and sends them to the sensing head 2. The electronic controller 3 also measures and records the current flowing to the CE 10 while the signal is being sent to the sensing head 2. The recorded data is then relayed back to the PC 4. When complete with one WE 12, the electronic controller switches to the next set of parameters or the next WE 12 and generates and collects further data. This is continued until measurements with each of the selected WEs 12 have been made. The number of signals sent to each WE 12 can also be set by the operator using the PC 4. With all measurements for each WE 12 complete that PC 4 will contain all the data in files for analysis.

The PC 4 applies artificial neural network models to the data. The data is obtained by measuring the current or voltage responses resulting from a set of voltage pulses applied between sets of electrodes. The data is characterised by statistical analysis. Data training sets are characterised in this way and used to create an artificial neural network model. This model is applied to new data sets in order to monitor or control the quality of liquids. This system is intended for use with both inline and bench top applications. Examples of applications include monitoring of inline cleaning processes, blending in a beer brewing process, and component concentration measurement in liquids.

Software of the PC 4 allows all of the required parameters to be set on the circuit, allowing a range of signals to be produced. The electronic controller 3 takes readings resulting from the applied pulses and sends this information back to the computer 4. The computer 4 software reads and stores these signal readings for analysis. The software analyses signals using statistical analysis techniques and has the ability to apply data to the neural network model. The computer 4 uses collected data, whose qualities are known, called a training set, to generate the neural network model. This model is saved and applied to new data in order to make readings.

A potential difference is applied between the RE 11 and the WEs 12. In order to maintain this potential a potentiostat is used. The potentiostat supplies charge (current) to the CE 10 so that the applied potential between the RE and the WE is maintained.

Many voltammetric techniques require that over time, a range of potentials be applied between the RE 11 and the WE 12. The types of signals used are "Staircase" (Fig. 3),

"Square Wave" (Fig. 4) and "Large Amplitude Pulse" (Fig. 5). After applying a particular potential, a measurement of the current flowing to the CE 12 is taken and recorded. Plotting the recorded current with the applied potential will give what is called a voltammogram.

This sensing system allows input potentials to the RE 11 and WE 12 to range from - 2.5V to +2.5V. The ranges can be varied to suit the analytes.

Handling of data is shown in Fig. 6. There is pre-processing of the raw data (for example, as shown in Fig. 10) to perform data reduction, and scaled data is generated to present to a trained artificial neural network (ANN) model. The model is applied to provide a quantified output.

Capture of raw data, generating a reduced data set for ANN training, and training the ANN is shown in Fig. 7, whereas run-time operation is shown in Fig. 8.

Pre-processing the raw data

The data generated by one electrode for a single measurement may contain many data points, perhaps 200 to 300 points. A single measurement will typically use data from more than one electrode and/or data generated using different parameters on a set of electrodes. Before the data is suitable for use with the neural network some preprocessing is performed to reduce the dimensionality of the data. The reduced data maintains relevant differences, or features, between data points. The method used for pre-processing data is based on computer image analysis techniques known as Shape Descriptors.

Shape Descriptors are a set of numbers that are produced to describe a given shape. Shape Descriptors for different shapes will be different enough that different shapes can be discriminated. The voltammograms used in this measuring system are essentially two-dimensional shapes containing a relatively large set of points. This measurement system uses Shape Descriptors as a means of reducing the voltammogram data.

The Shape Descriptors include: moments, area, orientation, compactness, and eccentricity

Moments are a descriptive measure of shape. Each voltammogram is a two dimensional shape. If a voltammogram consists of n points then the 0^th, 1^st and ^Λπd order moments can be calculated as follows where j = i%n, ( j = i module n):

0^th order moment

1^st order moments

m 10 ⁼ Σ - ^~£^{χl + χ}i^χ: ^{+ x}ϊ ^)(yt - ^yj ⁾

2^nd order moments

m 20 = Σ - 7z <Λ^{3 + χ}i ^χj ^{+ χ}i^χj^{2 + x}) )θ^; - yj ) i=l -¹ ^

m₂₀ = ∑—(^χ _t - X_j )(y- + yfy_j + y,y_j ^Z + y) )

m_xx ^{+ 2}yiyj + yj²) ^{+ χ}j (yf + 2y_iy_j + 3y_j ² ))

The 0^th moment, m₀Q is the Area of the shape. The 1^st order moments, m₁₀ and m_01; is the centroid of the shape. The 2^nd order moments give information on the distribution of the points of the shape.

The central moments are related to the general moments described above. It is more normal to use these for shape description. These are expressed as follows:

0^th order moment

1^st order moments

= u_m = 0

2^nd order moments

OT₁ ²O

M₂₀ = m₂₀

OT₀₁

M₀₂ = m₀₂

^oo m₁₀m₀₁

= m_n

The centroid is: m₁₀

X =

™₀₀

_ m₀₁ y =

™oo

Area: This is the 0^th moment.

Orientation: Using the central moments the orientation φ, of the shape (voltammogram) can be calculated:

Compactness: Is a measure of how packed a shape is: perimiter²/area. Eccentricity: The ratio of length of the longest chord in the shape to the longest cord perpendicular to it.

A combination of these shape descriptors is used to reduce the voltammogram data to dimensions more suitable for use.

Generating the Artificial Neural Network Model

This computer 4 uses a type of artificial neural networks (ANN) known as a Multi Layer Perception (MLP) network. MLPs use supervised training to generate mathematical models. In order to generate a model, data of known qualities is collected. The output of the MLPs is a set of values representing qualities of the liquid samples. This data set is called a training set and an outline of the training method is shown in Fig. 7. Important features of the training method are that this method is a systematic, "black box", way of constructing a model. Input to the ANN is data reduced as described above into only a series of numbers. There is only number for each of Shape Descriptor moment, area, orientation, compactness, and eccentricity. During the training phase data is continuously presented to the MLP. The training algorithm of the MLP uses resulting outputs and compares these to predefined outputs, and an error is evaluated. The algorithm uses this error to adjust the parameters of the MLP in order to reduce this error. This is continued until the error reduces below an acceptable level.

If the training data set suitably spans the working range of the inputs, a successful model will be generated, and the system may be calibrated to make a number of measurements.

Applying the Artificial Neural Network Model

Once the ANN model has been generated for an application, it is available for use.

New test data is collected by the electronic controller 3. This data is pre-processed and applied to the ANN model as illustrated in Fig. 8. Important aspects are that the ANN model is capable of providing values representing the input data. The system has the capability of providing information on a number of qualities of the fluids under inspection. The system can be readapted to analyse different qualities or different fluids.

Voltamrnetric sensors are also known as "electronic tongues", which is an appropriate title when they are applied to food. Here they are indeed used in the means of electronic tongues, and they discriminate between different samples of the same type of product. With this method taste maps are produced and the samples can so be differentiated, similar to the results obtained by analysis with human sensory panels.

Besides this application to food, the analysis system may also be applied to more specific applications such as monitoring of a single parameter in a mixture over time, or analysing a quality parameter in a process.

If the voltammetric sensor is to be applied in a full scale process, it has to be able to withstand a wide range of process parameters, regarding temperature, flow and pressure. In the harsh conditions of a cleaning-in-place procedure, the sensor system could, besides its use during the actual production, monitor the cleaning solutions, the cleaning progress and the cleaning effect. All the obtained data might also be recorded in the quality management systems that might be in place.

Application of the sensor system in a cleaning process step

In the process industry, and the food and pharma processing industries, CIP procedures are often operated. These can consist of various single cleaning steps with different parameters, and are mostly customised to the application. Different products need different cleaning methods. The voltammetric analysis system would be applied at the end of the process plant and would deliver an analysis of the process efficiency. The sensor head 2 is constructed to withstand harsh conditions and materials used, which would come into contact with the cleaning media, and has been tested in this situation. A surface design, which contributes to process flow profiles, also contributes to the usability of the sensor. Monitoring of cleaning processes is of importance within many industries. The cleaning effect has to be controlled. This could allow the cleaning time to be minimized, and so production time can be maximized. Further to this, the sensor system could be applied to indicate the need to start a full or partial cleaning procedure when the process plant shows major spoilage. Besides monitoring the cleaning effect and the strength and effectiveness of the cleaning agents, it is of great advantage to monitor the final rinsing procedure. Process plants in food or related processes are rinsed with food grade fresh water and when necessary steam sterilized. It is of great importance that in these final rinsing steps that no cleaning agents remain in the process plant. Many process plants, consisting of pipes, pumps, valves, and active machine parts are designed so that they may be cleaned without the need to take them apart. This is a major time and cost saving factor for the industry. Thus, all product-facing surfaces need to be rinsable and self-draining. This is also taken into account when designing the sensor front head. Only with this specification it can be assured that all cleaning agents can be removed after the cleaning. For the quality management system of a food processing facility it is important to have this parameter recorded. Heretofore, this has been done by a ph/conductivity measurement. The voltammetric sensor system can be applied here, as well used during the production, as a quality sensor on a desired process parameter.

Application of the sensor system in a blending; procedure

In some processes it is desirable to analyse and monitor a single component of a mixture and the system 1 could advantageously be used. The benefit of the voltammetric sensor system could here be found in a more complex application, for example the monitoring of several substances and their concentrations at the same time. This would be a benefit compared with the conventional method of using several sensor systems or even analyse in a laboratory after sample taking.

Example

A bench top experiment to measure the concentrations of citric acid and sodium chloride (NaCl) in aqueous solutions is as follows. A matrix of concentrations was prepared. These were labelled Cl, C2, C3, C4 and C5, to represent concentrations of citric acid, and Nl, N2, N3, N4, and N5 to represent the concentrations of NaCl in the prepared solutions.

Table 1

Mixtures of these two solutions were prepared with 5 graduations of NaCl and 5 graduations of Citric acid to create a matrix of 25 mixed solutions as shown in Fig. 9. These were named to reflect their content, for example N2C4 contained 6g/l NaCl and 4g/l Citric Acid. Fig. 9 illustrates the complete solution matrix of 35 solutions.

The objective was to use this 35-solution matrix to generate a neural network model that can be reapplied to similar test samples to determine concentrations of NaCl and citric acid. The modelling system accepts raw data in the form of voltammograms and outputs two values, one representing the NaCl concentration and the other representing the citric acid concentration in each test solution.

Each sample was measured in turn using the voltammetric sensing head 2 together with the electronic circuit 3. The readings were taken using the range of parameters on each of the working electrodes 12 Au, Ir, Pt, Rh. The measurements used stair case pulses (Fig. 3) in each case. Fig. 10 shows the raw data voltammagram output, relayed to the computer 4 for analysis. These measurements were repeated a further two times to give three sets of data in total. The first two data sets were used to train the neural network model and the final set was used to test the model

Before the ANN stage, the data was pre-processed in order to reduce the dimensions. This was performed by analysis of area moments of the voltarnrnogranis. These provide information about the geometric characteristics of shapes about a fixed point. This involves calculating the bounded area of each voltammogram, then the first order moments, which are the centroid of the voltammogram shape. Finally the second moments are calculated, and these are related to moments of inertia. These calculations were done for each voltammogram read from each electrode. These new values were then scaled, based on the spread of the data and the resulting values were used for presentation to the neural network.

The software used for both collecting and saving data, and generating the model was written in Matlab 6.0 and its neural network toolbox.

An example of voltammo grams from the Rh working electrode 12 is given in Fig. 10. This is the raw data captured from the sensing head 2.

The reduced data is shown in Figs. 11 to 22, in which

Fig. 11 shows a plot of citric acid solution concentrations against 0^th moment (Uoo) values for AuVl electrode-parameter set,

Fig. 12 shows a plot of NaCl solution concentrations against 0^th moment (Uoo) values for AuVl electrode-parameter set,

Fig. 13 shows a plot of citric acid solution concentrations against 0^th moment (Uoo) values for AuV2 electrode-parameter set,

Fig. 14 shows a plot of NaCl solution concentrations against 0^th moment (Uoo) values for AuV2 electrode-parameter set, Fig. 15 shows a plot of citric acid solution concentrations against 0 moment (Uoo) values for Ir electrode-parameter set,

Fig. 16 shows a plot of NaCl solution concentrations against 0^th moment (Uoo) values for Ir electrode-parameter set,

Fig. 17 shows a plot of citric acid solution concentrations against 0^th moment (Uoo) values for PtVl electrode-parameter set,

Fig. 18 shows a plot of NaCl solution concentrations against 0^th moment (Uoo) values for PtVl electrode-parameter set,

Fig. 19 shows a plot of citric acid solution concentrations against 0^th moment (Uoo) values for PtV2 electrode-parameter set,

Fig. 20 shows a plot of NaCl solution concentrations against 0^th moment (Uoo) values for PtV2 electrode-parameter set,

Fig. 21 shows a plot of citric acid solution concentrations against 0^th moment

(Uoo) values for Rh electrode-parameter set, and

Fig. 22 shows a plot of NaCl solution concentrations against 0^th moment (Uoo) values for Rh electrode-parameter set.

The results outputted from the ANN are summarised in Fig. 23. This plot shows the complete 35-point test data set, which has been processed by the ANN model. Each output result is a pair of vertically aligned square and round dots. For example, the left-most sample provides a result of zero NaCl and 1.5 parts citric acid.

Plots of the 35 test samples (prepared with the same concentrations as the training sample set,) are presented in Fig. 24 (NaCl) and Fig. 25 (citric acid). The plots clearly demonstrate the ability of the system to take voltammetric measurements and to translate these measurements into meaningful outputs. There is a spread about the "45°" line, but it is limited.

The invention is not limited to the embodiments described but may be varied in construction and detail. For example, the sensory head may not include a reference electrode. The counter electrode may be of a material other than stainless steel, such as a noble metal such as platinum.

Claims

1. An analysis system comprising: a sensing head having voltammetric working and counter electrodes, a reference electrode; a signal drive circuit for applying potentials across the working and reference electrodes when they are in contact with a sample; a capture circuit for capturing counter electrode current voltammetric response data, and a processor for processing said response data, wherein the processor comprises means for pre-processing the response data to reduce its dimensionality, for training an artificial intelligence model, and for applying the reduced data to the model to provide analysis results.

2. An analysis system as claimed in claim 1, wherein the processor pre-processes the response data by determining a shape descriptor to represent the response data.

3. An analysis system as claimed in claim 2, wherein the shape descriptor includes a value for voltammagram moment.

4. An analysis system as claimed in claims 2 or 3, wherein the shape descriptor includes a value for shape voltammagram area.

5. An analysis system as claimed in any of claims 2 to 4, wherein the shape descriptor includes a value for voltammagram orientation.

6. An analysis system as claimed in any of claims 2 to 5, wherein the shape descriptor includes a value for voltammagram compactness.

7. An analysis system as claimed in any of claims 2 to 6, wherein the shape descriptor includes a value for voltammagram eccentricity.

8. An analysis system as claimed in any preceding claim, wherein the model is a neural network model.

9. An analysis system as claimed in claim 8, wherein the neural network model is of the multi-layer perception type.

10. An analysis system as claimed in any preceding claim, wherein the data processor comprises means for training the model by successively applying signals across the working and reference electrodes, measuring resulting currents in the counter electrode, generating data corresponding to said measured currents, and applying training data to a transformation matrix.

11. An analysis system as claimed in any preceding claim, wherein the sensing head comprises the reference electrode co-mounted on a support also supporting the counter and working electrodes.

12. An analysis system as claimed in any preceding claim, wherein the counter electrode is centrally mounted, and the reference electrode is annular, surrounding the counter electrode.

13. An analysis system as claimed in claims 12, wherein the sensing head comprises a plurality of working electrodes.

14. An analysis system as claimed in claim 13, wherein the working electrodes surround the counter electrode and are between the counter electrode and the reference electrode.

15. An analysis system as claimed in claim 14, wherein the counter electrode is disc-shaped having a centre aligned with a sensor head axis, the working electrodes are disc-shaped and are equi-distant from the axis, and the reference electrode is annular and planar and is symmetrical about the axis.

16. A computer readable medium comprising software code for performing operations of the processor of the system of any preceding claim when executing on a digital processor.