1. The online eNose
Electronic noses, in short eNoses, are compact instruments comprising a sensor array that responds to
small variations in the reactive gas concentration in ambient air. Industrial gas emissions often hold a
certain amount of reactive trace gases that can be detected by eNoses. Since the released trace gases
can possess some odorous content the eNose has a potential to detect presence of industrial odours in
the environment. The eNoses are online connected to a remote computer system via a wireless data
communications link. Software on the remote computer interprets the signals of the remote eNoses in real-
time. A software agent automatically triggers an immediate alert when the eNoses detect a change of the
ambient air composition that has the likelihood of being malodorous.
The interpreting software features two basic functions. First it detects if the air composition is anomalous.
In case of a detected anomaly, the software compares the pattern of the actual eNose recording with a
database with reference patterns. The patterns are called the fingerprint of the eNose for the exposed gas.
The reference fingerprints are the result of training sessions where the eNoses are exposed to known
gases. By using a least square method the actual fingerprint is analysed against the available references.
If an appropriate fingerprint is available, then the odour type of the reference fingerprint is assigned to the
actual recording. The eNose apparently smells the fragrance of the reference. The quality of the
references is decisive for the smell recognition potential.
Although pattern recognition is an essential property of the eNose, the anomaly detection is perhaps even
more important. This compares to our own sense of smell. People can recognize familiar smells, but can
certainly detect unknown scents.
1.1 Training of eNoses and testing the system under real-life conditions.
Training of eNoses is a process to link other information about the gas situation during the recording of the
raw eNose data. Various training methods are available. Milan et al. (2012) reported a method in the Port
of Rotterdam by investigating time and place correlations between eNose recordings and odour complaints
of annoyed citizens reported to the 24/7 manned control room of the DCMR EPA. In Rotterdam this
training method is sufficient because there were over 30 eNoses permanently installed in the odour impact
area of the Port of Rotterdam for more than three years. Additionally, the DCMR EPA has developed
advanced methods for complaint registration and validation. Not many EPA’s have such a sophisticated
methodology. Neither has TATA Steel. Therefore the time/place correlation method between eNoses and
odour complaints was less appropriate. In this study a training method was performed to compare the
eNose recordings using dynamic olfactometry.
For training purposes gas samples were taken at the most relevant odour sources of the TATA Steel plant
in IJmuiden. The odour concentration of the samples was determined according to the European standard
for dynamic olfactometry, EN-13275, in the odour laboratory of VITO in Belgium. Next to the standardized
measurements, the sensitivity curve and odour threshold of the eNose for the different samples was
determined. The training resulted in the creation of a set of reference fingerprints of relevant TATA Steel
odours.
Events of odour annoyance in residential areas around the TATA Steel plant in IJmuiden were analysed
using this set of reference fingerprints. For this testing a dataset was collected that contains the complaints
of annoyed citizens, including the GIS-coordinates and timestamp, and the raw eNose data of the twenty-
five eNoses that were partially installed onsite the TATA Steel plant and partially in the communities. The
dataset also includes the wind direction recorded by five wind-vanes installed onsite the TATA Steel plant.
The dataset covers all recordings between Jan 1, 2012 and Dec 31, 2013. The time resolution of the
eNose and wind data is one minute.
1.2 Real-time monitoring
The eNose network is currently operational at TATA Steel. Each time one or more eNoses detect an
anomaly the system triggers an alert on a dashboard. The alert warns about the anomaly and refers to the
odour type of the best fitting reference fingerprint. The dashboard is online and updates automatically with
a one-minute refresh rate.