The main challenge of the gas sensor system is connection of the sensor to the energy management system and the radio system (Fig. 1), which should be as efficient as possible in order to achieve long service life, real-time capability, less maintenance, long-range capability, robustness, and mesh network capability.

For this purpose, a radio system was developed and evaluated, which uses a unique wake-up strategy to ensure that all sensors can be reached by radio permanently and in real time, but still has a very low energy consumption (Gamm et al., 2010). In contrast to current wireless communication systems (WLAN, Bluetooth, ZigBee), which minimise their energy consumption through periodic deactivation and activation (Kajikawa et al., 2016), this system allows for continuous real-time accessibility with low latencies. Since the individual sensor nodes also communicate with each other, measurement data can be recorded in star or multi-hop topology, if required. This offers the opportunity for an almost unlimited distance coverage in networks and opens the way for a significant reduction of costs. The radio system is robust, simple, efficient, maintenance free, and could be energy self-sufficient by using energy-harvesting technology. Even the radio protocol has been designed to be particularly robust and immune to interference for use in a safety-critical environment, where each measurement value is confirmed bidirectional at the receiver (Saez et al., 2017). The novel combination of these features with the low-power gas sensor enables the installation of the measuring system in inaccessible places as well as usage in a battery-powered mobile environment.

The second elementary component is the gas sensor. By using a MEMS-based hotplate, the power consumption of the sensor module (one active and one passive sensor) decreases to approximately 110 mW (Walden et al., 2015). This enables the use of the sensor in mobile applications. Because the suspended MEMS hotplate reaches the working temperature in less than 15 ms, a fast cycling or a fast thermal response is possible. This provides the opportunity for a deep sleep and a fast sensor wake-up, if necessary. The sensor is operated in a Wheatstone bridge to compensate disturbances such as airflow, ambient temperature, and changes in the thermal conductivity of the surrounding medium. One leg of the circuit is set by two 10 kΩ low-temperature-coefficient passive resistors. The other leg consists of two sensing MEMS pellistors. One pellistor is coated with the catalyst and the other serves as reference. A low-noise instrumentation amplifier amplifies the differential output voltage of the bridge.

In addition to the reduction of the power consumption by using a MEMS-based suspended hotplate, the working temperature, which is currently between 400 and 500 ∘C for commercially available sensors, could be decreased to 350 ∘C by the use of a novel catalyst material. Additionally, Walden et al. (2015) have characterised the sensor platform in detail, and they could show that lowering the working temperature by 100 K reduces the power consumption by 16 mW. The sensor requires about 71 mW at 450 ∘C and 55 mW at 350 ∘C. This reduces the power consumption by almost 23 %. The approximately linear behaviour in this temperature range depends on the dominant thermal conduction and the negligible thermal radiation. Commercial pellistors often use noble metal catalysts (Pd, Pt) dispersed on alumina supports with working temperatures exceeding 450 ∘C. This is necessary to reliably detect methane, which is the most inert combustible gas. In contrast to conventional pellistors, we used a non-precious metal oxide as catalyst. Tao et al. (2015) and Shao et al. (2017) described the use of NiCo2O4 as a good candidate for a complete methane oxidation in a temperature region between 300 and 400 ∘C. The gas-sensitive element consists of a low-temperature catalyst without using any precious metal. NiCo2O4 with a Ni:Co molar ratio of 1:2 was synthesised by co-precipitation according to the report by Tao et al. (2015), whereby the catalyst with a nanosheet morphology (Fig. 2, grey) was obtained as verified by scanning electron microscopy (SU-70 DEM). Subsequently, the NiCo2O4 catalyst was wet ground for 10 h in a steel jar with 2 mm steel balls to obtain an inkjet-printable material. The ground catalyst contains both the original nanosheets and the crushed particles (Fig. 2, blue), which should be beneficial to obtain a high surface area. A laser particle size analysis (LB-550, Horiba) confirmed the partial decreasing in particle size due to the grinding process of about 35 %.

The inkjet printing on the MEMS-based hotplate was performed with a DMP 2831 printer (Dimatix FUJIFILM, 2021). As shown in Fig. 1, only parts of the hotplate were covered with the catalytically active ink.

For the monitoring of flammable gases, a reliable detection of the lower explosion limit (LEL) is necessary. To report leakages or to evacuate people from harmful areas, 10 % of the LEL of methane should be detected. Here we present a gas-sensing device optimised for methane detection. Figure 3a presents the measured data for three different gas concentrations below the LEL of methane of 4.6 % (de Smedt et al., 1999). Figure 3b demonstrates that the sensor responds also to propane. The response to propane at the same concentration is about 18 % higher than to methane due to higher combustion enthalpy of propane. At a low working temperature of 350 ∘C (Fig. 3a), the sensor shows a stable baseline, and due to the high signal-to-noise ratio (SNR), concentrations far below the LEL of methane (<5 % LEL) could be detected. The measurements were performed in a gas-measuring chamber with a volume of 1 L and a gas flow of 250 sccm. A general description of the gas measurement station is given by Kneer et al. (2014). The ton time of the sensor was 500 ms with a sleep time of 3 s. Therefore, the duty cycle is about 14 %.

The sensor shows a response of approximately 5.5 mV to 8000 ppm CH4 in dry synthetic air. This is equal to a response of 6.9 mV per % CH4. Independent of supply voltages and mechanical boundaries, the literature describes sensor responses at higher temperatures. Xu et al. reported about 2 mV per % CH4 at 400 ∘C utilising 15 wt % Pd-Pt catalyst on γ-Al2O3 (Xu et al., 2011). The response of MEMS sensors with Rh2O3-Al2O3 catalyst at 400 ∘C is in the same range, presented by Su et al. (2013). As a result of lower working temperature, the probability that the pellistor will serve as an ignition source is substantially reduced (Papp et al., 2013). These gases were chosen because they are the common test gases for pellistors in industry. Methane is the main component of natural gas; propane is an important component of liquified petroleum gas (LPG) (Jessel, 2001).

After receiving qualitatively good results from the gas measurements, further optimisation of the battery lifetime was carried out. With a ton time of 500 ms for the gas sensor, the sensor node needs about 74 mAs of electric charge per measurement at a battery voltage of 3.2 V. This equals 237 mWs. The share of the gas sensor is 86 % (205 mWs) of the whole energy consumption. Only 14 % (32 mWs) is required for the microcontroller, the radio, and the peripheral circuits. This gives us the opportunity to measure more than 116 000 cycles with the built-in 2400 mAh (7.68 Wh) battery. With a sampling rate of 3 s, the system can work for more than 4 d. Commercially available portable pellistor sensors achieve several hours of measurement time to date. The sensor's battery lifetime could be further increased by setting the ton time to a smaller value. Due to the short heating time of the hotplate, the ton time could be set to 100 ms, which guarantees a stable supply voltage of the sensor (after 33 ms) and a readout of the analogue-to-digital converter (ADC). Figure 4 shows the difference in the energy consumption of the sensor system.

With a ton time of 100 ms, only 51 mWs are required to heat up the gas sensor. The energy consumption for the other parts remains constant at 32 mWs. The total demand of electrical energy can be reduced to 83 mWs, which is only 35 % of the initial value. A 2400 mAh (7.68 Wh) battery would now last more than 332 000 cycles. This equals an operation time of more than 11 d with a measuring interval of 3 s.

The runtime of one measurement cycle can be divided into four stages.

A: system start up with higher current of the voltage controller;

The sensor response is not affected by the duty cycle change of the system. The different sampling rates show almost the same results and raise time to the changing gas matrix. With this knowledge, it is possible to increase the battery lifetime by adapting the duty cycle factor. If the measurement interval is doubled, the battery lifetime increases by a factor of 2 as well. A measurement interval of 60 s leads to a sensor node runtime of more than 220 d and shows no loss of information for each single measurement compared to the measurement interval of 3 s. In this way, different work scenarios can be covered: a fast-repeating measurement cycle for safety applications or a long runtime mode for air quality sensing were less measurement cycles are required.

The humidity in the gas matrix is a major factor influencing the pellistor sensor response. Water molecules compete with the reaction partners for the catalytic surface, which reduces the active surface and thus the sensor response. In Fig. 6, the influence of three different relative humidity levels on the sensor response is shown. For these measurements, the ton time was set to 100 ms and the sampling rate to 3 s. The operation temperature was 350 ∘C.

The measurements results show that, at an exposure of 8000 ppm CH4 in synthetic air, the influence of a change in relative humidity is not significant. Within a relative humidity level change between 20 % and 60 % at 25 ∘C, the response decreases by 10 %, which corresponds in an equivalent methane concentration of 723 ppm. Only a slight baseline change of <0.5 mV for 20 % relative humidity could be observed after the gas treatment. The absolute sensor signal amplitude is in the same range as without having humidity in the gas matrix (cf. Fig. 3). Apparently, the non-precious metal catalyst material is not significantly affected by ambient humidity variations.