Malodour classification with low-cost flexible electronics

Abstract

Understanding body malodour in a measurable manner is essential for developing personal care products. Body malodour is the result of bodily secretion of a highly complex mixture of volatile organic compounds. Current body malodour measurement methods are manual, time consuming and costly, requiring an expert panel of assessors to assign a malodour score to each human test subject. This article proposes a technology-based solution to automate this task by developing a custom-designed malodour score classification system comprising an electronic nose sensor array, a sensor readout interface and a machine learning hardware fabricated on low-cost flexible substrates. The proposed flexible integrated smart system is to augment the expert panel by acting like a panel assessor but could ultimately replace the panel to reduce the test and measurement costs. We demonstrate that it can classify malodour scores as good as or even better than half of the assessors on the expert panel.

Fig. 1. Flexible integrated smart system.

a The high-level system architecture for armpit malodour classification problem. b Hardware blocks of the SRI and MLE used in the integrated FlexIC are interfaced to the sensor array.

Fig. 2. OFET devices as VOC sensors.

a Schematic of bottom gate, top contact OFET used in this work (S, D, and G are source, drain, and gate electrodes, respectively). b Compositions of the organic semiconductor layers used in the four different types of OFET (A–D). c Structures of all of the materials used in the OFET devices—low-k polymer is cross-linked P(BMA-r-MMA) and high-k polymer is P(VDF-TrFE-CFE).

Fig. 3. Sensor responses to various fabric swatches.

a Wet blank swatch fabrics. b–f Malodour swatch fabrics from different MMS values. A–D refer to four different types of OFET devices, and multiple sensors of the same type are used in the experiments.

Fig. 4. Operation modes and measurement protocol.

a Delta measurements to mitigate the sensor drift issue across different V_GS voltage and swatch modes. MB Delta refers to the I_DS difference between the malodour (M) swatch and blank (B) swatch modes at the same V_GS. BB Delta and MM Delta refer to the I_DS difference at the two V_GS voltages in the same mode. b The basic measurement protocol is shown for collecting the sensor outputs in the experiments where the wet blank swatch is used as a reference. First, the sensors are exposed to wet blank swatch and two readings of I_DS at two V_GS voltages (i.e., −3 and −3.5 V) are recorded. Then the sensors are exposed to the malodour swatch and two readings of I_DS at the same two V_GS voltages are recorded.

Fig. 5. Order Code Processor (OCP).

The components in OCP (digital SRI block) are shown. OCP supports 16 different instructions. There are three registers each of which has a 5-bit width. The fixed code sits in the programme memory where each instruction is microcoded in the microcode memory. The programme and microcode memory sizes are 256 and 832 bits, respectively.

Fig. 6. ADC measurement.

The measured ADC output is shown in 300 mV range at 1C offset with respect to the ideal ADC output.

Fig. 7. MLE hardware generation flow.

The stages for creating MLE hardware from the design space exploration (DSE) tool are shown.

Fig. 8. System prototypes.

a Die photos of the FlexIC integrating SRI and MLE (left), and OFET devices fabricated on a PEN substrate (right). The sensor array has eight sensors made up of four types of OFETs with redundant replicas. Only four sensors are used and powered on. b The high-level architecture of the demonstrator system. The blank and malodour swatches are placed into two vials. The headspace in vials is sucked into the tubes through the pump. The vial selection is controlled by a valve that allows only the headspace from the selected vial to be sucked into the main tube. The long tube carries the headspace onto the plastic e-nose sensor array. The sensor outputs are captured both by the microcontroller on the validation path and the FlexIC on the flexible integrated smart system. The FlexIC takes the sensor outputs and predicts its malodour score. The validation path uses the same e-nose sensor array but emulates the functions of the SRI and MLE in software and runs them on the microcontroller. This implies that the ADC in the microcontroller is used to digitise the sensor outputs from the e-nose sensor array but scaled back to 5 bits. The predictions from both systems are displayed on the LCD screen. c Picture of the demonstrator box—the valve and pump are not shown in the pictures, and the microcontroller is located on the backside of the main board. The swatch vials are not part of the box and are connected through the tubes to the box.