Calibration
All sensors show a degree of variation when compared to each other. This can be the result or the manufacturing process, the way in which the sensors operate or a whole range of other causes. Some expensive sensor units have built-in electronics that detect and compensate for these variations, but that is not always possible. Calibration is the process of comparing the output from a sensor with one or more known reference values so that any differences can be taken into account.
When measuring the electrical conductivity (EC) of a nutrient solution, for example, your sensor will give you a value, but without a reference for comparison, you do not know how accurate that value is. However, if you compare the value from the sensor with one taken by an electronic meter such as the one shown below, you can adjust your sensor reading so that it matches. For pH, the reference values would typically be provided by samples of known pH.
Linear adjustment
The simplest strategy is to assume that the relationship between the reference values and the measured values is linear. That means that if you collect a series of measurements with their matching reference (true) values, you can plot them on a graph and the pattern will be roughly a straight line. Let's use the table of values shown below as an example.
| Measured value (mV) | Reference value (μS/cm) |
|---|---|
| 0 | 0 |
| 200 | 183 |
| 240 | 448 |
| 490 | 803 |
| 640 | 1080 |
| 860 | 1490 |
| 1080 | 1930 |
The measured values are shown in mV because that is the signal obtained from a typical analogue sensor, and the true values are shown in microsiemens per centimetre (μS/cm) which is a typical unit of electrical conductivity.
A scatterplot can be created from this data where the measured values lie on the x-axis and the reference values lie on the vertical axis. The result is shown below. As well as the points themselves, the graph also includes a trendline, or line of best fit. There are mathematical procedures to work out where the line needs to be, but if you are using a spreadsheet application it will create the line for you. Once you know the line of best fit, you can work out its equation in the form
where m is the slope of the line, and c is the offset in the vertical dimension. The equation for the line is also shown on the graph.
Once you know the slope and offset, you can take any measured value and work out the true value my multiplying by m and adding c. Note that in this case, c is negative so adding c is the same as subtracting the absolute value of c.
An interesting thing to notice in the example data above is that although the measured value that corresponds to a reference value of zero is also zero, the line of best fit does not pass through the origin of the graph. There are several options for what to do in such a case. The node uses the simple solution of accepting that there might be a slight inaccuracy at measurements close to zero because that gives the best fit over the majority of cases.
If you are feeling adventurous, please consult this thorough guide to best practice for producing calibration curves.
Polynomial adjustment
Linear calibration is used by the node to adjust the output of the pH sensors. However, it was found in testing that a second-order polynomial curve (quadratic) was a better fit with the data from the EC sensors. This is where the line of best fit is defined by an equation such as
The table below shows some actual data collected from the EC sensors during testing. The following graph shows a polynomial best fit line with the associated equation calculated by the spreadsheet.
| Measured value (mV) | Reference value (μS/cm) |
|---|---|
| 0 | 0 |
| 442 | 724.31 |
| 795 | 1068.11 |
| 846 | 1171.72 |
| 876 | 1151.56 |
| 1130 | 1352.71 |
| 1150 | 1469.64 |
| 1340 | 1380.10 |
| 1720 | 1574.65 |
| 1790 | 1427.64 |
| 2060 | 1597.22 |
| 2080 | 1446.17 |
