Humboldt-Universität zu Berlin, Institut für Informatik

Lehr- und Forschungsgebiet Signalverarbeitung und Mustererkennung High Resolution Temperature Measurement Technique for Materials Sciences Experiments in Space (Copyright © 1994 by the International Astronautical Federation. All rights reserved. Document No. IAF-94-J.2.234)

Abstract

General terms of reference and global concept

     Figure 1: Measuring device TEGRA at the crystal growing furnace CSK - 4

Figure 1 shows a general arrangement, provided for experiments in crystal growth under the conditions of microgravity in the Russian spacecraft MIR. Primarily the data are expected to be available for the scientific evaluation, but it also has a direct influence on the further experimentation [1], [3].

The data provide the basis of the answers to the following questions:

• How exact were the thermal conditions and timing of the experiment ?
• Which absolute temperatures and temperature gradients exist at chosen points and times in the probe?
• Has any acceleration (mechanical vibration) occurred that is relevant to the experimental results?
• Have any special conditions occurred which make it necessary to cancel the experiment and restart it?
• Does the experiment meet the original aims, i. e. have any new effects been detected?

Therefore, the measuring device TEGRA should be performed for the precise measurement of the absolute temperature, high-resolution measurement of temperature differences and the simultaneous measurement of microgravitation (µg) values. The arrangement of different modules was made on a functional basis, and structural results are shown in figure 2.

     Figure 2: Modular structure of the measurement device TEGRA

• The temperature measuring module (called the temperature module) conditions the data delivered by the experimental probe such that it can be processed by a microprocessor. This means amplification, filter and analog-to-digital conversion.
• The digital module performs the processing of a freely programmable temperature scan rate and transmits data via a serial interface.
• The microgravitation measuring module (called acceleration module) simultaneously scans three inputs. Each of them can be connected to an accelerometer.
• To perform a fail-safed power supply with high efficiency, a special supply module was developed.

The specific requirements for these modules and their technical realization are described in the following sections.

General problems and concepts for the technical solution of the temperature module

Background and general requirements

1. Design of equipment that allows differential measurements for all connected thermocouples (differential thermometry) as well as absolute temperature measurement. The effective range should be adaptable to the voltage range of type K (NiCr-NiAl) or type N, and type S (Pt-PtRh) thermocouples, such that the maximum resolution can be attained. Within each probe, all thermocouples should be of the same type.
2. Any probe socket should be designed for up to five thermocouples. The quality of the insulation between the wires and the sheath should not affect to the data acquisition. Generally, a minimum of crosstalk and disturbance effects should be achieved by outside dc and ac currents in the sheath. Therefore, all inputs should be designed as really differential amplifiers with very good common mode rejection also in the RF-range.
3. Design of a changeover switch for measurement points that allows a high channel separation (greater than 100 dB). It must have such a low error (noise, offset voltage and error currents) that it is below the resolution of the entire system.
4. Development of a measurement amplifier with a low random disturbance and a stable gain. The requirement is to achieve a temperature resolution for the differential thermometry of 0.01 K, which corresponds to a voltage resolution of 0.1 µV when using Pt-PtRh-thermocouples.
5. Utilization of an A/D-Converter (ADC) with as large a conversion width as possible. We require a temperature resolution of 0.01 K in a measurement range of 300 K to 1500 K, which gives an A/D- conversion width of 17 bits. In the case of earlier devices for the experiment TES, fewer requirements resulted from the separation of absolute and differential temperature measuring channels. However, it requires a fixed allocation of thermocouples for the difference measurement.
6. For measurement of the absolute temperature at the junction of thermocouples, the comparison temperature must be known. Thus, a special sensor has been assigned to this measurement. The maximum resolution of the signals received from thermocouples depends on several independent factors.

Basically, a limit is reached if the input signal is less than the effective sum of noise voltages and disturbance on the input of the first amplifier. A further limit is the quantization error due to the finite resolution of the ADC’s.

By using digital signal processing algorithms (the simplest case is average), the quantization error can be reduced under certain conditions. This corresponds to an effective increase in the resolution. However, such a procedure has practical limits. They are resulting, for example, in a deviation of the noise from the normal distribution and the non-calculable drift.

Development of the electronic concept

The maximum reduction of disturbances can be achieved by using a measurement amplifier with symmetrical inputs because the thermocouples have a symmetrical structure. For example, this problem can be solved by using an isolation amplifier for each input. However, the tolerances of offset and gain values of presently available amplifiers do not guarantee the necessary accuracy for differential thermometry. In the previous projects [4] [5], only non-symmetrical inputs were realized. The cross-coupling effects between several thermocouples should be reduced. Practical measurements on sheathed thermocouples of the NiCr- NiAl type yield a low insolation resistance of about 10 kOhms at temperatures of ~800 K. This causes a temperature measurement error in the region of 0.1 percent. Furthermore, it was not possible to differentiate between common and differential mode disturbances in previous TES experiments.

To ensure the full accuracy of the measurements, we need thermocouples with better insulation properties than available at the moment. So we have proposed a new method in TEGRA to allow a free combination of thermocouples for the differential measurement. The main idea is to connect all thermocouples with separated capacitors in the basic state. This yields a high impedance between the thermocouples and the preamplifier also among the thermocouples themselves. Figure 3 shows this principle with two thermocouples. Theoretically, a full distortion suppression will be expected even by a single defect of low-impedance connection between the thermocouples and their respective sheaths.

     Figure 3: Schematic diagram of the SC-circuit for measurement inputs

After leaving the basic state, the concerning capacitor will be connected only to the input of the preamplifier.

This is why the common mode disturbances of the thermocouples ideally cannot reach the preamplifier. The accuracy achieved depends on the relationship between on- and off- resistance (and the values of storage and dispersion capacity) of the analog switches. From practical measurements, a noise reduction of up to 120 dB (which corresponds for a voltage ratio of 1/106) can be derived over a wide frequency band. For noise suppression, a larger value of capacity gives a better result. However, there exists an upper limit. A low pass filter is performed by the capacitor in conjunction with the on-resistance of the analogue switch additional to the internal resistance of the thermocouples. The corner frequency should be not too low. For instance, if one measurement is taken every second, the time constant of the low pass filter should not be greater than approximately 100 ms. With a total resistance of 1 kOhm, this allows a maximum capacitor of 100 µF. In this application it is possible to use a simple Operational Amplifier (Op-Amp) instead of an instrumentation amplifier. The requirements of this preamplifier are pointed in the following.

The first important view is the input bias current. It must be sufficiently low, else the voltage at the capacitor will change according to

(1)

For example, we need a voltage change of less than 0.1 microvolts. Measurement durations of 100 ms and an input current of 100 pA require a capacitor minimum value of 100 microfarads. An additional limit is caused by the input current through the on-resistor of the multiplexers. That is why an Op-Amp with FET-inputs should be used preferably. Otherwise, it is necessary to compensate the current error.

Secondly, we consider the offset voltage and its drift parameters. Using a simple calculation, the summary offset errors can be compensated. Therefore, an additional reference channel similar to the measurement channels must be provided. A zero-voltage is taken to its input. This allows a partial correction of errors caused by the input current, too. In the result of practical measurements it could be assumed that an input bias current of 1 nA is uncritical for this operation.

However, all the drift parameters (and the different channel properties) determinate the remaining voltage error Duoffs. For example, a measurement duration of about t = 0,5 s and a voltage error of s < 0,1 microvolts caused of operating temperature fluctuations of DT = 5 K/min result in the greatest permissible drift value of

(2)

This can be achieved by precision Op-Amp’s easily.

Thirdly, the noise behaviour of the preamplifier is very important for the highest resolution. Further elements of the input circuit (multiplexer and feedback resistors) must be designed to minimize the noise. Considering the frequency range, we find the lower corner frequency at the sampling rate of 1 Hz. The upper limit of 100 Hz can be performed by a lowpass filter in the hardware (the preamplifier itself), also in software procedures. The centre frequency amounts 10 Hz. Noise density values at this frequency are often given in usual data sheets. Figure 4 shows the least noisy Op-Amp’s now available.

     Figure 4: Typical electrical properties of
     selected Operational Amplifiers

To obtain the RMS noise voltage, the integral of the noise density function (flicker noise by 1/f- characteristics) must be solved over the interesting band width. For the considered case, this voltage (in nV) is approximately five times higher than the given density value at 10 Hz. For example, the best FET-input type (Burr-Brown’s OPA 627) generates effectively noise of about 75 nV RMS. Resulting peaks limit the voltage resolution. The noise voltage can be reduced by using an Op-Amp with bipolar inputs. But its input bias current and the resulting current noise are the reasons that no better results can be established.

Basically, the white noise of all the resistors acts additional. The formula

(3)

gives the well-known resistance noise voltage, which limits the sensitivity and resolution fundamentally. For example, the RMS voltage of 30 nV will be generated at room temperature by a summary resistance of

(4)

if there are no further noise sources. Available switching circuits attain typical values as low as 50 Ohms. The additional noise, caused by the input bias current, must often be determined experimentally. Noise coefficients for analogue switches are not declared. The share of noise of feedback-resistors can be avoided by using metal-film resistors. Currently it could be assumed that 0.1 microvolts are the resolution limits for measurement with thermocouples. This is attainable when current values are lower than about 1 nA. We achieved best results with the MAX 400 circuit. In the sum a resulting noise is about five times higher than the theoretical limit when an absolutely noiseless Op-Amp is assumed.

To select an ADC (Analog-to-Digital-Converter), first the class of integrating converter, e.g. dual-slope, should be considered. Resolutions up to 18 Bits (and up to 22 Bits by multiple combinations) are possible, but the conversion time is about 100 ms. To measure 16 points, it takes more then one second.

An alternative is the class of Successive-Approximation-Converter. About 16 Bits are received in conversion times of below 100 microseconds. Well-designed LSI-circuits minimize the additional hardware. The practicability of oversampling has decided the matter to select this kind of converter. Increasing the resolution is possible when a procedure is reducing the differential linearity error. In the result, realistic values range up to 18 Bits in a total time of 10 ms.

A theoretical consideration yields e. g., that a quadruple lot of non-self-correlated conversions allow to increase the resolution of another Bit. A necessary condition is that the effective noise quantity (s-value) is essentially greater than the resolution limit indicated by the LSB and caused by converting process or amplifier stages. In practice the increase from 15 to 18 Bits requires a minimum of 64 conversions, more secure are 128 conversions. Measured results are in good agreement with the theoretical expectations.

There is an important difference between the maximum output voltages of different types of thermocouples. To use the full range of ADC, the preamplifier should be fitted with a built-in stage for switching between two gain values. The digital module automatically considers the topical gain value.

The main structure of the temperature module is fixed now. Concerning the availability of specific electronic circuits, some details of the development and engineering process are characterized as follows:

• A measurement of the thermocouple’s comparison point (room temperature) is performed by using special sensor devices (AD 592, manually selected to accuracy of + 0,25 K). The input circuit of this channel could be simplified because a galvanic separation is not attainable. By special adaptation of the measurement range, a temperature resolution of less then 1/100 K is reachable (important for differential measurements).
• Two different gain values should be provided to adapt the output voltages of the used thermocouples. In the present solution, the voltage ranges amount to 50 mV (basic gains without wire bridge), 12.5 mV respectively. The data output format is fixed by 0.1 microvolts in both ranges. In the 12.5-mV-range, the amplifier input noise limits the practical sensitivity, but in the 50-mV-range the resolution of the ADC limits the accuracy of the practical results. A wire connection between two pins of the probe plug is necessary to select the higher gain for Pt-PtRh-thermocouples.
• By fixing the number of inputs, an important fact was that 8-to-1-analogue multiplexers with low leakage currents and on-resistances of less then 50 ohms are available. The differences of thermocouple voltages inside the circuits are neglectable. In the result, the solution is a double construction of eight-channel preamplifiers concerning two probe sockets. A reference channel is established in both circuits.
• The realization of two preamplifiers with separated sockets is important for applications, which require two separated probes with different voltage ranges.
• The association of the eight inputs to channels are given as follows:

  1       internal ground, reference for input 2
  2       compare point temperature measurement
  3       zero-voltage reference for all measuring inputs 4 to 8
  4       thermocouple 1
  5       thermocouple 2
  6       thermocouple 3
  7       thermocouple 4
  8       thermocouple 5
  

• Generally, it is possible to perform the differential measurement regime simply by modifying the input circuit if a pair of charged capacitors would be switched together and generate the difference voltage directly. However, the noise and disturbance properties of all electronic devices limit the input sensitivity to about 0.1 microvolts. So there is no advantage by modifying the input circuit. The voltage differences are calculated in the TEGRA digital module. The same resolution is reachable and a free association between all the five thermocouples of the probe is possible.
  A further advantage resulting in the high temperature resolution of > 1/100 K is to take a thermocouple reference to its own previous value. This is important to watch differential temperature fluctuations at a fixed measurement point. In this case, there is no fundamental distinction between absolute and differential temperature measurement.
  

At this point, all the important aspects of concept and development of the temperature module are discussed. In summary, figure 5 shows the whole structure of all the block elements.

     Figure 5: General view of the block structure of
               the temperature module

Digital Module

• ADC controlling and data collection
• Preprocessing depending on a user program
• Communication with the Crew-Interface-Computer CIC

The main data storage would be done by the Crew-Interface-Computer CIC, which is connected via a serial link.

Temperature Module Controlling

1. Generation of the scan rate points
2. Non-overlapping switching of the storage capacitor at the input circuit
3. Multiplexer control for channel selection
4. Consideration of the settling time of pre- and main amplifier
5. Control of Analog Digital Converter
6. Collection of digital measured data
7. Calculation of corrections

• The maximum rate of a temperature increase in the probe is 50 K/s.
• The maximum rate of the ambient temperature in a working phase is about 5 K/min.
• The range of measurement is 300 K ... 1500 K.
• The repetition rate of a measuring cycle is 1 s.
• The temperature resolution is 0.01 K.

To perform a continuous operation of the analog module and to avoid thermal transient phenomenons, it is favourable to run one measurement cycle every second. The measurement program selects the number of data that have to be saved. The structure of this measurement program should be optimal adapted to the CSK-furnace. Since the CSK program is divided into a maximum of 60 steps, there are up to 99 steps provided in the TEGRA program. Each step includes the storage and transmission of several data sets, which were sampled at equidistant times. It is possible to select the number of measuring points within a data set. The scan rate has a constant value of one second. Because of an analog settling time of approximately 5 ms and a conversion time of 50 µs, an oversampling rate of 128 is possible. This reduces the noise and the virtual resolution of the Analog-Digital-Converter increases. (See section 2). So the total conversion time is approximately 20 ms. A full data set which includes all measuring points will be obtained in less than 0.5 seconds.

Controller and communication functions

The following requirements should be considered:

• The warranty of a parametric programmability of the steps concerning start and deceleration times, to the sample rate and to the selection of relevant measuring points,
• non-volatile storage of all parameters,
• possibility of change during the running program step,
• no loss of collected data in case of disturbance or short dropouts of power supply,
• deadlock free operation also for longer disturbance phases or longer power break downs,
• galvanic decoupled communication channel for avoidance of influences, stressing the analogous assemblies because of peripheral devices (ground loops), and
• an additional, redundant communication channel with a program independent transmission, including all measured data with a constant sampling rate of one second.

In earlier experiments (ARP, TES) the process of measuring programs was described with a certain number of measurements (so events). This approach was simple and reliably attainable. The RTC could be renounced. However, a disadvantage exists if there were a loss of power or mishandling, the time covering to the experiment will be lost. Either the series of measurements had to start again, or the measurement was carried on without knowledge of its downtime.

In contrast to this, a time control guarantees a fixed reference of measuring data with the real time. Possible interruptions do not have any influence on it. Another advantage is, that a measuring program will be started variously:

1. Immediate start of the first step after the power up.
2. A delayed start at a programmed date. It is independent from other actions and allows a fully automatic operation.
3. Start at a prior date and inception of a series of measurements not in the first step but in any of the current time corresponding steps. This is important particularly in case of power fail or manual interruptions of the program.

     Figure 6: Hardware components of the digital modules
               and its communication with the CIC

The communication of the digital module with the CIC is based on a serial interface. It works accordingly to the RS-232 standard and performs a galvanic separation from the other modules.

Its most important jobs are

1. the actual transmission of measured data, and
2. the receiving of new measurement programs or some single step instructions within the program.

The data format is configured such, that

• the default setup of 9600 baud, 8 Bit, no parity can be used preferably,
• there are no special claims to the CIC with respect to the reaction speed, and
• no special claims on the calculator regarding real time behaviour are necessary. For the storage of acquired data, a simple program on the CIC should be installed. For example, the MS-DOS command "COPY COM1: filename" is enough to save data in a file.

The measurement program consists of up to 99 free programmable steps, each of which encloses individual start and stop dates, the measurement rate and the selected channels. Any information of the measurement program will be transferred from CIC to TEGRA by an arbitrary number of steps. The steps are stored in one or more files that can be transmitted in any order.

Especially the transmission of a new measurement program is possible while the actual measurement program is operating. TEGRA will finish the actual program and start the new one if its starting time is reached. In particular it is possible to deliver a new measuring program while already running a program. Using an internal buffer memory for the temporary storage of data, a high level of reliability will be achieved. On interest of a high data security, within the CIC it is preferably suggested to use a block organized file structure. TEGRA supports this preceding by generation of experiment-specific and step-marked filenames. These filenames will be placed in the head of the transmitted measurement data. In case of a CIC program crash, only the existing data of the actual step will be lost.

A second serial interface (called the front interface) is assigned. It transmits all the measurement data independent of running program and state of the CIC-Interface. This output is redundant and indicates the general operation. By connecting a terminal to the front interface, a special TEGRA monitor program will be running instead of the regular measurement program. This makes it possible to access the internal resources (e.g. all memories) for testing and variation of the TEGRA-internal software.

Some aspects of realizing the acceleration module

By using the accelerometer QA-1400, the technical conditions to perform a proper interface are given. The following requirements determine, for the most part, the technical structure of the device:

• The possibility of the connection to three accelerometers, preferably for orthogonal measurement of three-dimensional components of acceleration.
• Minimum sample rate of 200 Hz for each of the three channels.
• The guarantee of a dynamic range greater than 60 dB.
• Full suppression of DC components (important under normal earth conditions).
• Single adjustment of the measurement range to the typical operation conditions (e. g. a range of 40 mg with the resolution of 10 µg).
• Interface for data output to CIC without compression.

     Figure 7: Block diagram circuit of the
               acceleration module of TEGRA

Due to the simple conditions and the very large integration scale, the quantity of components is reduced quite a lot. Consequently, the analog and digital components can be placed on a single printed circuit board. The digital controlling problems are similar to the temperature measurement. Therefore, the hardware solution is similar as well.

In the first version of the operation software, the processing of measured acceleration data is reduced to very simple procedures. The collected data can be stored into a buffer for a maximum of about one minute. Then they must be transmitted to CIC via RS-232 interface that uses the XON/XOFF handshake protocol.

The installed operating program on the CIC could be easily modified for special problems. For example, there is enough space for detection of peaks or Fast-Fourier-Transformation. It is possible to store all the collected data without compression. The direct installation of suitable procedures by using gained experiences is foreseen for later developments. The hardware concept very well meets the requirements for this.

Technical concept of power supply and implementation details

By development and design of the power supply module it is necessary to develop and simulate with particular systematics. This way is described briefly as follows:

1. All supply circuits have to be electrically separated. The electrostatic insulation is necessary wherever a separated circuit is connected to any external device (e. g. probe, supply mains, CIC). In the temperature module, the high resolution ADC requires two separate supply circuits for analogue and digital function blocks in standard applications. However, if the currents for compensation of different potentials between grounding points are neglectable, the supply circuits can be linked.
2. The galvanic separation is realized by using a transformer in a switching supply converter. The switching frequency and its harmonics are effectively between all the terminals of the same and different windings. Therefore, some frequency-specified decoupling elements are needed. Commercial types of switching converters often neglect this aspect. The result is a high distortion voltage between several supply circuits, and the voltage converter is not practicable for high resolution measurement.
   

   Figure 8 shows the effect of this distortion voltage (Usw) on the measuring input circuit. It is realistic to assume a voltage source in a series with a capacitor to describe the switching disturbances.
   

   

        Figure 8: Simplified modelling circuit of the switching DC-to-DC-converter
                  to show the effect of the disturbance voltage
   

   It is possible to minimize the capacitance csw, but the most common converters do not care about this. To optimize the supply converter, two steps are used:
   

   • A special winding technique to minimize the cross-coupled voltages, and
   • an additional compensation capacitor between the primary and the secondary circuit is eliminating the rest of disturbances.
     

   For example, a common converter with Ck = 150 pF and Usw = 9 V could be improved to the developed converter with Ck = 55 pF and Usw = 0,3 V.
   

3. The next step is the determination of all needed voltages and currents for all the separate modules and supply circuits. Under specific conditions in the TEGRA device, we need the following voltages:

   Secondary supply voltages and currents:

   1. temperature module +/- 5 V 10 mA
   2. heating for reference voltage 15 V 150 mA
   3. digital module (temperature) 5 V 250 mA
   4. analog parts of the acceleration module +/-15 V 50 mA
   5. digital part of acceleration module 5 V 250 mA
   6. power supply for RS 232- Interface 5 V 100 mA

4. Considering the number of the separated voltages and the low quantity of disturbances, the whole structure of the supply module must be fixed. In the present design, a push pull voltage converter could be equipped with eight step-down switching regulators. This arrangement guarantees a high degree of effectiveness (about 80%) and a very high disturbance suppression.
   

5. In order to achieve good conducted susceptibility and to minimize the conducted current emissions, some filter circuits are necessary to ensure the full measurement accuracy. On board the spacecraft MIR, the general operational instructions and test requirements also must be observed. Further details see [4].
   

Summary

The acceleration measurement module was designed for general applications preferably in µg-environments. All the accumulated data will be transmitted to the CIC, where it is possible to install various evaluating programs and the data store.

Some problems concerning design and implementation of the power supply module are discussed, and some selected technical parameters are presented.

References

[1]   A. Bewersdorff, G.P. Görler, R. Willnecker, K. Wittmann, R. Kuhl,
      R. Röstel, M. Günther, G. Kell:
      MEASUREMENTS OF HEAT CAPACITY IN UNDERCOOLED METALS
[2]   J.P. Granier, Y.Dancet, P. Faucher, S. Riaboukha:
      MICROACCELEROMETRE EXPERIMENT MIR MICROACCELERATION CHARACTERIZATION
[3]   C. Barta, A. Triska, J.Trnka, L.L. Regel:
      PROCEEDINGS OF THE 5TH EUROPEAN SYMPOSIUM ON MATERIAL SCIENCE UNDER 
      MICROGRAVITY; ESA SP-222, 413 (1984)
[4]   M. Günther, M. Karl, G. Kell, R. Willnecker, F. Winkler:
      TES-ELEKTRONIK - EIN RAUMFLUGTAUGLICHES TEMPERATURMEßGERÄT FÜR DIE 
      DIFFERENZ-THERMOANALYSE UND DIFFERENZ-KALOMETRIE
[5]   R. Kuhl, H. Quaas, H. Süssmann:
      ARP - A MULTIPURPOSE INSTRUMENTATION FOR EXPERIMENTS IN MATERIALS
      SCIENCES IN SPACE, Preprint, 37th congress of IAF, 1986

Contacs

Director:      Prof. Dr. Beate Meffert
Developed by:  Manfred Günther, Lothar Heese, Gerald Kell, Thomas Morgenstern, Frank Winkler
               Humboldt-Universität zu Berlin, Institut für Informatik
               (in cooperation with DLR Köln; founded by DARA Bonn)

To the Home-Page of professorship "Signalverarbeitung & Mustererkennung" at Institut für Informartik