Hydrogen Sensor for Oil Transformer
Applied Nanotech, Inc.
Abstract – A hydrogen sensor for detection of H2 dissolved in
transformer oil has been developed for the use in a stand-alone
dissolved gas analyzer (DGA) which will also assess the relative
humidity saturation of oil. The sensor uses palladium
nanoparticles as a sensitive material that is selective to hydrogen.
The DGA will be capable of measuring dissolved hydrogen in
concentrations from 50 ppm to 4000 ppm.
Keywords - hydrogen sensor; palladium nanoparticles; oil
transformer health monitoring; dissolved gas analysis
Hydrogen is a major fault gas observed in oil filled power
transformers and is an indicator of possible corona discharge
between the windings of the transformer coil. Timely detection
of a change in concentration of dissolved hydrogen is critical in
monitoring the transformer health and can prevent the
transformer failure. There are a number of commercially
available techniques currently used for the detection of
hydrogen dissolved in transformer oil. They are based on gas
chromatography, use of metal oxides, thermal conductivity,
and fuel cell type combustible gas sensors . All the
techniques are based on the measurement of hydrogen
concentration in a sample of gas obtained from oil rather than
directly measuring H2 in oil. Also, most of these technologies
are either high cost or suffer from the issues of cross-sensitivity
to combustible gases other than hydrogen.
It is known that palladium (Pd) metal can be used as
sensitive material for detection of hydrogen in air. Palladium is
intrinsically selective to hydrogen, since H2 readily dissociates
on the Pd surface and diffuses into Pd while changing the Pd
lattice constant and forming palladium hydride. This process is
often called a phase transition in palladium. However, at
standard pressure and temperature, the phase transition takes
place at rather high partial pressure of hydrogen , around
1% in air, that is not practical for dissolved gas analysis.
The sensor developed by Applied Nanotech, Inc. (ANI) uses
Pd nanoparticles to lower the H2 partial pressure at which this
transition occurs. The surface effect in which hydrogen atoms
occupy subsurface sites  prior to “bulk” phase transition
makes it feasible to detect very low hydrogen concentrations.
The “effective” phase transition in Pd nanoparticles is a
smooth rather than a step function of hydrogen pressure ,
and allows detection of hydrogen in the range spanning over
several orders of magnitude.
ANI has exploited the unique property of Pd nanoparticles’
interaction with hydrogen at low concentrations in the
miniature hydrogen sensor that is ideally suited for H2
monitoring in power transformers. ANI is developing this
hydrogen sensor as a part of a dissolved gas analyzer in
partnership with a major supplier of equipment for energy
utilities industry. The DGA includes temperature and humidity
sensors for a broader assessment of the transformer operating
conditions and will also be capable of operating with both
nitrogen-blanketed and free-breathing types of transformers.
The ANI design approach uses a network of Pd
nanoparticles deposited on a resistive substrate. An
electroplating technique is used for Pd deposition in a
chronoamperometric mode, where nanoparticle size and
interparticle distance are controlled by the proper choice of
plating parameters. Fig. 1 shows an SEM image of
nanoparticles deposited on a substrate. An average nanoparticle
size is approximately 30 nm, and the gaps between the
particles appear to be on the order of 1 nm or less.
In the presence of hydrogen, Pd nanoparticles undergo a
phase transition, increasing in their size and creating electrical
contacts between each other. The change in the electrical
resistance of the nanoparticle network is thus a function of
hydrogen concentration (Fig. 2). Narrow gaps between
nanoparticles ensure good sensor sensitivity to hydrogen
concentrations as low as 10 ppm.
Fig. 1. SEM image of Pd nanoparticles deposited on a sensor substrate
The sensor in its current design has a glass substrate and is
mounted onto an 8-pin TO-5 package. Fig. 3 shows the sensor
having both a working and reference elements; their terminals
are wire-bonded to the package pins. The reference element
keeps track of the changes that can be detected by a working
element but not related to the presence of hydrogen, such as
temperature deviations or any long-term drifts. In the standard
configuration, the sensor is mounted on top of a miniature TEC
module such that the temperature of the sensor can be
controlled with good precision using an integrated feedback
The palladium metal is a catalyst that promotes the reaction
of hydrogen oxidation and formation of water molecules .
This means that the amount of hydrogen dissolved in
palladium also depends on the concentration of oxygen
present. During the catalytic reaction, oxygen takes away
hydrogen atoms from the surface of palladium nanoparticles.
In the DGA design, we use ambient air with the known oxygen
The DGA sampling system consists of two sampling
volumes: one for air and one for oil, separated by a gas
separation membrane. Changing the sampling volume ratio, it
is possible to adjust the sensor sensitivity span for H2 in air to
the range of concentration of hydrogen typically expected in
transformer oils. This approach allows measuring
concentrations of dissolved hydrogen from 20 to 5000 ppm,
which extends beyond the range of interest for this application.
For the consistency of the measurements, stability of the
sensor operation, and due to the fact that at different
temperatures the amount of hydrogen that can be dissolved in
oil is different, the oil sample as well as the hydrogen sensor
are kept at a fixed temperature of 80°C. Keeping the sensor at a
fixed temperature using a PID algorithm helps significantly
simplify the sensor calibration at the production stage as well
as easing the spot check calibration process in the field. Since
at fixed temperature the Ostwald (partition) coefficient for an
oil-hydrogen system is a constant value, the sensor can be
calibrated in an air-hydrogen mixture.
III. EXPERIMENTAL RESULTS
The sensitivity range of the sensor spans over several orders
of magnitude. The sensor is typically sensitive to the H2
concentrations in air to as low as 10 ppm. The response of the
sensor changes monotonously up to the concentration of 1%
and higher (see Fig. 4), and saturates at nearly 4% H2 in air.
The sensitivity curve shown above can be fitted fairly well
with a logistic function. This function describes initial
exponential growth of some parameter, followed by the
inflection in the growth curve and finally exponential
saturation. The logistic function best fits the sensitivity curve
for a hydrogen concentration in a log scale. The equation for
the normalized response R (drop in the sensor resistance) can
be written in the following form:
where RMIN is the normalized minimum resistance achieved at
sensor saturation, C is the hydrogen concentration, C0 is the
inflection concentration, and a is a power coefficient
describing the curve slope at the inflection point. The inflection
concentration is the hydrogen concentration at which the
exponential growth of the sensor response to increasing
Fig. 3. Hydrogen sensor mounted on a TO-5 package.
Fig. 2. Phase transition in Pd nanoparticles results in the detection of H2
Fig. 4. Sensor response to hydrogen in air
hydrogen concentration changes to sensor saturation. The
power coefficient is related to the phase transition miscibility
gap: the higher the power coefficient, the narrower the
miscibility gap. For a bulk Pd the miscibility gap is very
narrow, and the sensor based on large particles would operate
like a hydrogen-driven switch. A broad gap (low a) indicates
that particles of palladium have very small dimensions, as well
as rather wide distribution of the nanoparticle sizes.
The physics and chemistry of Pd-H interactions in the sensor
is quite complex, and even a good fit with an analytical
function does not yield readily a simple model for hydrogen
sensing effect. However, this approximation can be useful for
the sensor characterization and calibration. For example, in
case of consistent distribution of nanoparticle sizes, which
gives us an estimate for a power value a, that can also be
considered a constant, sensor calibration can be done only at
two points – one at saturation point in order to determine RMIN
and the other one near the inflection point, from which the real
position of the inflection concentration C0 can be easily
determined. As far as this approximation works well in the
semi-log scale, it can be used for sensor calibration over
several orders of magnitude of concentration values.
The sensor response and recovery times (t90) in hydrogen-air
mixtures are usually below 10 seconds (see Fig. 5) for low and
high hydrogen concentrations. However, at concentrations near
100-300 ppm, we observe an increase in the response time that
can be as long as several minutes. Interestingly, this feature
coincides with the inflection point on the concentration curve.
This may be explained by a transition from “surface”
sensitivity to “bulk” sensitivity, where two processes compete
at intermediate H2 concentration.
The nature of such a competition can be clarified if we take
into account an observation that the sensor saturates at much
lower hydrogen concentrations in the lack of oxygen. In pure
nitrogen-hydrogen mixtures, the sensor will max out at H2
concentrations in the low ppm range. This means that oxygen
plays a crucial role in the phase transition in nanoparticles.
Most likely, the hydrogen trapped in a loosely packed sub-
surface Pd lattice reacts with oxygen at a rate high enough to
deplete H2 in the bulk. When oxygen is not present, hydrogen
quickly diffuses into the bulk resulting in the phase transition
even at very low H2 concentrations. The bulk-to-surface
gradient in the packing density of Pd atoms may be taken into
account as well, though it should not lead to any dramatic
changes in the outcome of such a model. Thus, we can assume
that at the inflection concentration, there is a balance between
the diffusion rate and the water formation rate.
In the transformer DGA, however, the response time is
limited by the hydrogen diffusion throughout the oil sample
and the area of the gas separation membrane. In most cases, the
response time of the sensor during the measurements in oil is
less than 30 minutes.
The sensor design is rugged and withstands wide variations
in ambient temperature and atmospheric pressure. It has water-
proof housing and is designed to be easily mounted on an oil
drain valve at the bottom of the transformer oil tank. The DGA
has an RS-232 user interface and current loop outputs for
hydrogen concentration (in ppm by volume) and relative
humidity saturation (in %) values. In addition to this, water
concentration in the units of ppm (by weight), as well as the oil
temperature (in °C) values can be directly obtained via RS-232
port. The embedded firmware enables the user to read the
measurement status, change constants and calibration data, and
also perform field calibration of the sensor if necessary.
A Pd nanoparticle hydrogen sensor for detection of H2
dissolved in transformer oil has been developed for the use in a
stand-alone dissolved gas analyzer (DGA) which will also
assess the relative humidity saturation of oil. The sensor can
be used for many other applications where detection of
hydrogen at concentrations below the lower flammability level
(4% H2 in air) is required. These include leak detection,
hydrogen storage, hydrogen production, fuel cells and other
industrial and automotive applications.
 T. Cargol, “An Overview of Online Oil Monitoring Technologies,” 4th
Weidmann-ACTI Technical Conference, pp. 1-6, San Antonio, 2005.
 F.A. Lewis, The Palladium Hydrogen System, New York: Academic
 C. Oh, S.-G. Lee, C.-H. Park et al., “1H NMR Study of Hydrogen in
Palladium Nanocrystals”, J. Korean Phys. Soc., vol. 43, pp. L958-L962,
 M. Suleiman, N.M/ Jisrawi, O. Dankert et al., “Phase Transition and
Lattice Expansion during Hydrogen Loading of Nanometer Sized
Palladium Clusters”, J. of Alloys and Compounds, vols. 356-357, pp.
 G. Pauer, A. Winkler, “Water formation on Pd(111) by reaction of
oxygen with atomic and molecular hydrogen”, J. Chem. Phys., vol. 120,
pp. 3864-3870, 2004
Fig. 5. Hydrogen sensor response to 1000 ppm H2 in air.