Tsoil test for CME project

General Comments:

The first field test of the intelligent sensor array, ISA, will take place during the CME project, May through July of 2004. A total of 72 Tsoil measurements will be made. The test will consist of 3 individual arrays each array having 4 microscale nodes and 1 mesoscale node. The microscale nodes will be built around the Crossbow Mica2 dot board. Each Mica2 will ingest data from 6 temperature probes, LM20 chip. The data from the Mica2 will be transmitted to the mesoscale node, BitsyX board. Data from the BitsyX will be sent to the main data acquisition system at the base of each tower.

 

 


Initial Tsoil Tests:

The Tsoil probe will be interfaced to the Mica2 via a 10 bit A/D. There is concern that the A/D will not give the accuracy needed. Based on numerical calculations averaging a minimum of 10 points will give an accuracy better than +/- 0.25C. We are now running tests in the oil bath to confirm our calculation.

Test Procedure:

Two Mica2 units were tested. Each had 6 LM20 temperature probes interfaced to the A/D. On unit #1 the temperature probes were mounted on a small circuit board while unit #2 had the probes attached directly to the A/D via wires. The reason for this test was to help determine which approach would provide the best coupling to the soil, ease of construction, and ease of calibration. Each probe was dipped in an electrical coating material to prevent moisture damage (this was not necessary in the oil bath because the oil is non-conductive). Each probe was inserted in a glass tube (unit #2 probes were all together in one tube) which were then inserted into the bath to the same depth as the reference probe. The glass tubes were filled with oil to improve coupling to the bath oil. The selected temperature profile was from 50oC to -30oC in steps of 5oC. A minimum soak time of 90 minutes was used at each point. During the last 10 minutes of a soak data were collected from the Mica2s and the reference. The data from the Mica2 consisted of the 6 temperatures and a battery reference signal.

Analysis:

The first part of the analysis was to determine a transfer function for the raw data to engineering units. In the past a simple polynomial function has worked well with this probe.

Plots – Mote1Mote2 : These show the results of a simple linear fit. The y-axis, residuals, is the difference between the fitted data and the reference. Analysis of higher order polynomials did not significantly improve the fit. These fits did not take into account any changes in the battery voltage. The circuit mounted probes, Mote1, show a slightly better fit than the directly wired probes.

Plots – Mote1N & Mote2N & Mote2N' : These plots show the result when the raw data is normalized to the battery reference, Traw/battery. A simple linear fit was used again. There is a significant improvement in the residuals. The anomaly at the 35oC point in Mote1N has not been resolved. Applying a 2nd order fit to the Mote2 data gave better residuals, plot Mote2N'.

Plots – Mote1avg & Mote2avg : These plots show the results if we average 10 data points based on the 2 fits for each mote. The approach was to apply the fit(s) to the data then do the 10 point average.

 

Conclusion:

Based on these initial tests, the averaging and normalizing of the data will be needed to achieve the desire accuracy. A second oil bath run is now being done. At some point we will need to test the stability of the A/D over temperature.

 

Oil Bath Test #2:

The second oil bath test had similar results. The fits were not as good but there was no major variations in response from the first test. One of the LM20 probes on Mote2 failed during the test. Also the anomaly point at 35oC on Mote1 during the first test decided to move to 40oC.

Plots - Mote1 , Mote2 , Mote1N , Mote2N' , Mote1avg , Mote2avg

 

Note: After this test it was determined that the output signal from the motes did not represent the true raw counts. The software was applying a simple conversion from raw counts to millivolts. The code was changed so that we would get raw count information. The following tests were done in this mode.

 


Temperature Stability Tests:

This test was done to determine if the Mica2 boards had a dependency on temperature changes. Each channel of the Mica2 had a fixed reference voltage applied while the board was subjected to a temperature change from 50oC to -30oC in steps of 10oC. A soak time of 30 minutes was used at each temperature point. The test was conducted in theThermotron chamber which has a typical response time of 5oC per minute. The reference voltage was set to 1.5000 volts and was monitored with a Keithley 2000 DVM. All data was collected by the ClaLab PC running LabView. Only the Mica2 boards were in the chamber.

Results:

The results were good and bad. The good part is the A/D of the Mica2 does not have a temperature dependence. This can be seen in the first plot of figure ADtest . The bad news is the reference voltage device, Vref, does have a temperature dependence (refer to plot 2 in ADtest). This is a problem because of the method used to convert from raw A/D counts to volts. The A/D uses the battery supply voltage as the voltage reference signal to the A/D. In other words, Vin = Vin(counts) * Vbat/1024. Since the Mica2 can not monitor the battery voltage, which can change over time, it dedicates one A/D channel to monitor a reference voltage, Vbat = Vref * 1024/Vref(counts). As the battery voltage changes the counts for Vref will change appropriately and Vbat can be computed. However, the temperature dependence of the Vref device will effect the computed value for Vbat. Plot 3 in ADtest shows the computed Vbat using the Vinput signal as a reference. The battery voltage is stable over temperature. The last plot in ADtest is the computed Vbat using the actual reference signal. A possible solution to this problem is to monitor the temperature then apply a correction to Vref with respect to temperature. The Mica2 dot does have an on-board thermistor that could be used.

Temperature Applied Correction to Vref:

A second temperature test was conducted with two additional items, monitor the on-board temperature and change the supply voltage, Vbat, at each temperature point. A temperature correction could be computed for Vref assuming the following criteria: (1) the A/D does not have a temperature dependence, (2) the temperature probe is dependent on the battery signal, and (3) the change in the Vref counts with respect to temperature is not dependent on the Vbat level. The previous test satisfies the first condition. The second condition was verified by examining the circuit board layout. The temperature probe is in series with the 10K resistor tied to Vbat. As Vbat changes, the voltage input to the A/D from the temperature device will change proportionally; ie, the output counts from the A/D for temperature will not change due to changes in battery level. The final condition was verified based on the new data set collected. FigureADtest2 shows 2 plots. The first is the changes in the Vref counts due to temperature and Vbat changes. The Vbat went from 2.7 to 3.3 volts in .2 volt steps. More important is the second plot showing the changes in the temperature probe counts. The plot shows no significant effects in the changes in Vbat.

The Correction Steps:

The first step was to apply a Stienhart/Hart fit to the response of the on-board thermistor. The second step was to compute a correction value to apply to Vref as the temperature changed. This was done by selecting an arbitary temperature point, using the Vref counts at that point, and subtracting all other Vref counts at other temperatures to determine the change with respect to temperature: Vdiff = Vref(tx) – Vref(20C). A fit was then applied to the the Vdiff data using the computed temperature: Vdiff = A0 + A1*Tboard + A2 *Tboard^2.


 

Another Option:

Due to the temperature dependence of the Mica2 Dot, tests were run on the standard Mica2 to see if its reference source had a temperature dependence. The results were minimal to the point that no temperature correction was required. Because of this we decided to change to the standard Mica2 board.

Raw Data to Calibrated Data:

In order to eliminate changes in the battery voltage, Vbat, which is directly tied to the A/D, the reference voltage is monitored along with the other A/D/channels. A ratiometric calculation can then be made which removes any fluctuations due to Vbat. To improve overall measurements each Mica2 Vref needs to be determined accurately. This was done by applying a known voltage,Vknown, on an A/D channel and collecting raw counts for this channel and the Vref channel. The Vref value can be computed by: Vref = Vknown * Vref(count)/Vknown(count).

Once the Vref value is known the same ratiometric approach can be applied to the raw Tsoil counts: VT = Vref * Traw(count)/Vref(count).At this stage a fit can be applied to VT.

CME04 Post Cal Results

Plots were not created of the pre-project calibrations; however, post project cal plots do exist. These can be seen at Tsoil Report.