Field test evaluation of instrumental odour monitoring systems with a novel in-situ calibration approach

49 004

   Instrumental Odour Monitoring Systems (IOMS) are employed for continuous objectified technical olfaction. However, the reference for odour perception is the human nose, surpassing technology with outstanding sensitivity and selectivity for many compounds. In order to achieve comparable results, IOMS have to fulfil manifold requirements, resulting in considerable effort in engineering and cost. One-size-fits-all solutions are unlikely to function properly but are nevertheless marketed as the resulting performance is difficult to assess.

   In the course of the publicly funded research project SEPEG (sensor networks for the objective perception of odour sources, BMBF FKZ 01IS17087) all aspects of a complete IOMS solution are scrutinized in their respective context. A main focus of the project is on field test evaluation in real installation situations. Two installation sites have been equipped with networks of real-time IOMS devices.

 Wolfhard Reimringer1*, Julian Joppich2, Martin Leidinger1, Thorsten Conrad1, Bettina Mannebeck3, Christoph Mannebeck3, Andreas Schütze2
1. 3S GmbH, 66115 Saarbruecken, Germany
2. Laboratory for Measurement Technology, Saarland University, 66123 Saarbruecken, Germany
3. Olfasense GmbH, 24118 Kiel, Germany
*wolfhard.reimringer@3s-ing.de

 

   Competing interests: The author has declared that no competing interests exist.

   Academic editor:  Carloz N. Díaz

   Content quality: This paper has been peer-reviewed by at least two reviewers. See scientific committee here

   Citation:  Wolfhard Reimringer, Julian Joppich, Martin Leidinger, Thorsten Conrad, Bettina Mannebeck, Christoph Mannebeck, Andreas Schütze, 2021, Field test evaluation of instrumental odour monitoring systems with a novel in-situ calibration approach, 9th IWA Odour& VOC/Air Emission Conference, Bilbao, Spain, Olores.org.

   Copyright:  2021 Olores.org. Open Content  Creative Commons license. It is allowed to download, reuse, reprint, modify, distribute, and / or copy articles in olores.org website, as long as the original authors and source are cited. No permission is required from the authors or the publishers.

   ISBN: 978-84-09-37032-0

   Keyword: IOMS, electronic nose, dynamic olfactometry, validation method

Download Presentation

Video

 

Abstract

   Instrumental Odour Monitoring Systems (IOMS) are employed for continuous objectified technical olfaction. However, the reference for odour perception is the human nose, surpassing technology with outstanding sensitivity and selectivity for many compounds. In order to achieve comparable results, IOMS have to fulfil manifold requirements, resulting in considerable effort in engineering and cost. One-size-fits-all solutions are unlikely to function properly but are nevertheless marketed as the resulting performance is difficult to assess.

   In the course of the publicly funded research project SEPEG (sensor networks for the objective perception of odour sources, BMBF FKZ 01IS17087) all aspects of a complete IOMS solution are scrutinized in their respective context. A main focus of the project is on field test evaluation in real installation situations. Two installation sites have been equipped with networks of real-time IOMS devices.

   In CEN TC 264 WG 41, a standard has been developed for the quality assessment of IOMS. This requires a reference data set with representative variations of source and background. The key question of comparability to olfactory standards is pursued by a novel method of in-situ calibration with dilution series of source samples, the odour concentration of which has been determined via dynamic olfactometry as per EN13725:2003.

   In three training/validation campaigns, the forced calibration strategy has been implemented and tested for technical and commercial viability. The results show that the general approach is promising, while providing impetus to further development of the methods, algorithms and the IOMS used in the experiment.

 

1. Introduction

   Machine olfaction is often connected with the term “electronic nose” which is well suited for advertising but lacks the necessary modesty regarding the well-known shortcomings of sensor systems or analytics (Boeker 2014). In order to emphasize the seriousness of current efforts, CEN TC 264 WG 41 coined the term “Instrumental Odour Monitoring System” (IOMS). In this working group, validation strategies have been laid out to enable verifiable manufacturers claims, as the term “instrument” designates a measurement device with known accuracy. Validation procedures are distinguished for absence/presence, odour identification and odour concentration.

   One important aspect of the validation methods is the representativeness of the calibration data set. All application-specific variations must be covered, including source variations (odour concentration and composition), environmental parameters and background variation. In the past, training and calibration experiments have been carried out (Reimringer 2017, Schmid 2017) that relied on natural variation over a certain period. While this approach needs long lead time due to slow seasonal fluctuations and sparse odour events, the collection of odour reference data is particularly challenging. Grid inspection information from an EN16841-1 survey can deliver absence/presence data only and must be carefully synchronised with the IOMS response. Thus, forced application of prepared odour samples seems a viable alternative to overcome lack of odour concentration data and to allow for a brief validation phase in IOMS deployment.

   In this work, a practical implementation of an IOMS deployment and validation phase has been carried out as a field test experiment. Technical and commercial viability have been examined in order to identify requirements for further IOMS development and input for the standardization process.

 

2. Materials and methods

   The sample preparation process for the forced calibration method is similar to dynamic olfactometry according to EN13725. A source sample is taken and examined for its odour concentration. Acquisition of the source sample involves the usual risks, in the case of the field test sites these are height (20 m stack at industrial exhaust air treatment plant) or drowning (open water bodies in a sewage treatment plant). For each source sample, a background sample is taken in the vicinity of the IOMS installation site. The background sample is checked for absence of the respective target odour via direct evaluation as dynamic olfactometry is not applicable near the odour threshold.

   For sample preparation, the sampled background air is split up into a set of n secondary sampling bags, n-1 of which are then injected with a calculated volume of the source sample to obtain a pre-defined odour concentration. The remaining sample bag contains the background without target odour. Direct evaluation of the diluted samples ensures that the target odour is present.

   The IOMS used in the field tests is a 3S EnvironmentalCheckerOutdoor (ECO), which can sample odours from up to three inlets via a pump. The ECO can be modularly equipped with various sensors; for odour assessment, a module with four metal oxide sensors has been used. One inlet sampled the ambient air through a tissue filter (to prevent dust and insects entering the system) and one inlet was used to connect the sample bags for calibration (cf. Fig. 1). Calibration mode could be activated via an interface extension, sampling an exact volume from the sample bag and tagging the associated raw data sets.

   Algorithm development employed supervised learning methods. For odour concentration, partial least square regression (PLSR) has been used. The input for the underlying model is aggregated from the sensors’ response collected in the raw data sets, initial dimensionality reduction is performed by means of feature extraction (mean values and slopes of the temperature cycled gas sensors).

   For the performance claim, CEN TC 264 WG 41 suggests the use of Chebychev’s inequality as there is a small set of data with unknown statistic distribution. The form of the claim is given by the key performance parameters x and α, where the IOMS output lies within the confidence interval (1/x ∙ REF) ≤ IOMSout ≤ (x ∙ REF) for a significance level (probability) of α. REF indicates the “true” value from a reference measurement. Thus, an instrument with x = 4 and α = 0,95 would give a measurement output with an odour concentration between one quarter and four times the reference value for 95% of the measurements taken and would therefore match the performance of dynamic olfactometry itself (cf. Boeker 2007).

 

Field test devices. Left: Installation in an impacted residential area.

Fig. 1: Field test devices. Left: Installation in an impacted residential area.
 Right: Sample bag attached to inlet B of the sensor system

   Three training campaigns were conducted, each covering both field test sites. All sample sets were applied to all respective IOMS (8 devices in the industrial exhaust scenario, 4 at the wastewater plant). The whole process of sampling, preparation and application took about 5 hours for one sample set, employing two teams in offset shifts and a mobile olfactometry lab. In effect, each campaign took one working week to accomplish. The target odour concentrations were 0, 5, 10, 25, 50, 100, and 200 OUE/m³ in the first week, while 10 and 200 were omitted for the following campaigns.

 

3. Results and discussion

   Fig. 2 shows training results of one IOMS after the first campaign. The sample bags were applied to the calibration input for three minutes at a controlled flow rate of 500 mL/s. This resulted in four usable measurement cycles from the metal oxide sensors per sample. The majority of the data was used for algorithm training which turned out to be more successful for the industrial exhaust application compared to the wastewater application.

   Detailed analysis of the data revealed a significant problem with the background contained in the sample bags which dominated the sensor response in the form of a strong offset (blue vs. cyan signals in Fig. 3). Initial assumptions of this being an effect of the switched inputs could be ruled out by the application of same samples to both inputs.

   The calibration data set from the samples mentioned above revealed considerable deviation between “true” value and IOMS prediction. In the form proposed in the CEN working group, a confidence factor x = 15,5 at a significance level α = 70% was found (exemplary data from IOMS #13, industrial exhaust scenario). This is certainly not a set of parameters to be advertised but the result of shortcomings in training and calibration, worsened by the pessimistic estimation of the Chebychev approach.

Usage of the diluted samples for algorithm training.

Fig. 2: Usage of the diluted samples for algorithm training.
Left: Regression model for industrial emission. Right: Wastewater treatment.

Comparison of background air and bag responses.

Fig. 3: Comparison of background air and bag responses.
Left: Response to bag samples at calibration input (0, 50, 5, 25, 100 OUE/m³) and ambient input (100).
Right: Discrimination (PCA) of ambient air, bag zero and 100 OUE/m³ sample at both IOMS inputs. Colours indicated in the PCA plot also apply to the exemplary cycles on the left, indicated as vertical lines in the quasistatic signal and overlayed in the “selected cycles” plot.

4. Conclusions

   From the work described above, a significant on-site effort becomes apparent. In the scope of this project, three sampling campaigns were conducted, with an overall financial effort in the range of a one-year EN 16841-1 odour survey. This becomes prohibitive for commercial projects that do not necessarily need real-time odour monitoring. Furthermore, an IOMS cannot be put into service immediately by this method but needs a significant lead time for training and validation, preventing fast response to acute problem situations. Thus, calibration strategies have to be developed for relevant setup or “factory calibration” in order to obtain systems that can deliver relevant information out of the box, taking into account possible on-site improvements of the algorithm as well as the validation of the odour prediction according to the emerging standard.

   Despite due diligence in sample preparation, the zero offset of the sampling sets remains a yet unsolved problem. On the one hand, the odourless background air must be representative for the IOMS site, on the other hand, this complicates the sample preparation and heavily increases expenditure: With installation sites that are always more or less affected, a representative background cannot be obtained at the original site. With IOMS networks, each installation site would need its special background sample. In combination, this questions the original idea of a forced calibration that rules out the uncertainty of occurrence for a representative set of situations to be validated. In urban impact sites, the variation of non-target background situations is likely to scale up the required extent of such a set far beyond the current minimum of 8 sample pairs. Future work has to clarify how the requirement of an absolute odour concentration calibration can be fulfilled for strong background variations.

   An obvious finding is that the discrimination between target odour(s) and background must be improved on. As the definition of such targets is application specific, this cannot be achieved under all circumstances. For some components, a bioreceptor-like performance could be engineered within the sensor system itself. In general, however, a strong background situation constitutes a highly unfavourable signal-to-noise ratio. Evaluation methods including neighbouring nodes and data history are likely to improve data quality to some extent but must be complemented by considerate placement of the sensor systems. Therefore, further work will be undertaken on all three aspects in real application scenarios.

   Another approach that has been ideated during the current field tests but not yet implemented is the combination of source and impact measurement. For this, an IOMS is equipped for emission measurement and installed complementing an impact site IOMS network. This could support network detection methods and provide interesting insights in the validity of dispersion calculation. A consistent response of the IOMSs deployed is crucial, therefore calibration strategies will have to be designed including an initial factory adjustment.

 

Acknowledgement

   The work described above has been partly funded by the German Federal Ministry of Education and Research (BMBF) under FKZ 01IS17087.

 

5. References

   Boeker, P. 2014. On ‘Electronic Nose’ methodology. Sensors and Actuators B 204, 2-17, ISSN 0925-4005. doi: 10.1016/j.snb.2014.07.087

   Boeker, P., Haas T. 2007. The measurement uncertainty of olfactometry. Gefahrstoffe –Reinhaltung der Luft 67, 331-340.

   Reimringer, W., Conrad, T., Schütze, A. 2017. Citizens Network as Reference for Odor Impact Sensors - a Case Study, Proc. ISOEN 2017 - The International Symposium on Olfaction and Electronic Nose, Montreal, Canada, May 28-31, 2017, Proceedings p. 179-181.

   Schmid, S., Mannebeck, B., Hauschildt, H. 2017. Odour complaints in urban environment. 7th IWA on Odours and Air Emissions, September 25.–27., 2017

   EN 16841-1:2016. Ambient air – Determination of odour in ambient air by using field inspection. Part 1: Grid method

   EN 13725:2003. Air quality – Determination of odour concentration by dynamic olfactometry

All the content here under Creative Commons license