2
nd
Conference on “Monitoring & process control of
anaerobic digestion plants”
Leipzig – March 17-18, 2015
Electronic Nose Technology for Reactor State and Biogas Quality Assessment in Anaerobic Digestion
Gilles Adam
1, Sébastien Lemaigre
2, Xavier Goux
2, Philippe Delfosse
2, Anne-Claude Romain
11University of Liège, Arlon Campus Environnement, 185 Avenue de Longwy, B-6700 Arlon, Belgium
2Environmental Research and Innovation Department (ERIN), Luxembourg Institute of Science and Technology (LIST), Rue du Brill 41, L-4422 Belvaux, Luxembourg
The anaerobic digestion (AD) process (syn. biomethanation) has a high potential to contribute to sustainable energy production because it is one of the most advanced options available to convert especially wet biomass into multipurpose
fuels (CH4and H2), valuable products to serve the green chemistry/biorefinery sectors, and fertilizers readily available to
agriculture. Nevertheless, a major limitation to further development of the sector is the difficulty to keep the anaerobic flora of the digesters in optimal conditions of activity. Numerous methods have been assessed to properly monitor the process but none appears to be ideal [1].
Electronic noses (e-noses) have been documented as potential tools to be employed in the AD domain, especially for process monitoring [1]. E-noses could also be used for biogas quality assessment and for safety purposes (biogas leak detection, biogas combustion, etc.) These instruments obviously present a potential in AD field but have also clear limitations. A previous work by Adam et al. [2] demonstrated that an e-nose could detect different disturbances in lab-scale reactors with an exclusive focus on the gas phase of reactors.
INTRODUCTION
MATERIAL AND METHODS
Lab-scale continuous reactors:
Overfeeding campaigns E-nose and classical process state indicators: continuous data collection
Data analysis and Model development This work presents the application of an e-nose evaluated on a pilot-scale continuously stirred tank reactor (CSTR) for two purposes: i) process stability monitoring and, ii) biogas quality assessment (methane and hydrogen sulphide content). Both purposes are essential factors for the economical viability of biogas plants. An important aspect of this work is that one instrument, an e-nose composed of 6 metal oxide semiconductor (MOx) gas sensors, was tested to reach both purposes.
AIMS of the STUDY
0 1 2 3 4 5 6 7 8 9 10 0 10 20 30 40 50 60 70 80 90 100 110 120 O L R ( g / L / d a y) , b io g a s a n d C H 4 ra te ( L / L / d a y) Time (days) OLR
Methane production rate biogas production rate
Disturbed reactor Collapsing reactor 3 4 5 6 7 8 1.0 1.5 2.0 2.5 3.0 p H N H 4 ( g / L ), A lk . ( m l C O 2 e m it te d ) Alkalinity N-NH4 pH
•Inoculum: sludge from the anaerobic digester of a waste water treatment plant (mesophilic conditions). •Substrate: dried sugar beet pulps.
- 4 continuously stirred tank reactors (CSTR) 100l
- mesophilic temperature range (37°C)
Overfeeding campaigns E-nose and classical process state indicators: continuous data collection
Parameters in the digestate:
Parameter Units Measurement method Measurement Frequency
Biogas specific
production l/h drum-type wetgas meter hour-1
CH4concentration % (Volume) Non Dispersive
Infra-Red Sensor (NDIR)
(2 hours)-1
CO2concentration % (Volume) NDIR (2 hours)-1
H2concentration (Volume)ppm Metal Oxide Semi-conductor (MOx) with molecular sieve (2 hours)-1 H2S concentration ppm (Volume) Electro-Chemical Sensor (ECS) (2 hours)
-1 Parameters in the biogas:
Parameter Measurement method Measurement Frequency
pH
Saturated calomel electrode
day-1
Total solids (TS) Gravimetry (VDI 4630) week-1
Volatile solids (VS) Gravimetry (VDI 4630) week-1
Total alkalinity (TA) (BiogasPro, Volumetric Germany) week-1 Ammoniac Nitrogen (NH4-N) Volumetric (BiogasPro, Germany) week-1 Model development
RESULTS and DISCUSSION
1) Process stability monitoring 2) Biogas quality prediction
E-nose configuration and measurements (on reactor n°3)
•Array of six commercial MOx sensors •Sensor chamber maintained at 50°C •Every hour:
- 1 min gas sampling - 9 min sensor exposure to diluted
biogas - 50 min purge with clean air •Dilution ratio of 1:25 for biogas:air mixture •Data recorded every 30 s
Steps:
1) Acquisition of the stable response (µS) for each sensor 2) Normalize data
3) Standardize Data
4) PCA Model training using 30 first days 5) Normalize and standardize new observation 6) Select PC number
7) Calculate T² and Q²
8) If T² and Q² < control limit: update mean, standard deviation and correlation matrix
9) Return to step five
1) process control 2) gas quality RMSEP: 1.92% RMSEPCV: 1.99% RMSEP-Test: 3.7% R²: 0.64 R²cv: 0.61 %X explained: 90.81% %Y explained: 64.07% 45 50 55 4 5 5 0 5 5 CH4, 3 comps, validation p re d ic te d
↑ Organic Loading Rate (OLR): VDI 4630
20 min sensor response x 2 records/min x 6 sensors = vector of 240 data/observation
Model: Kernel PLS (using PLS toolbox, R 3.1.1. software) Model training:
Data set of the 30 first days (300 observations) Model testing:
10 fold cross-validation Test set: 138 data (day 31 to 60)
finally: testing on the complete data set (day1 to 110)
Phase I Phase II 1000 1500 2000 2500 3000 3500 4000 50 0 1 00 0 15 0 0 2 0 00 2 5 00 3 0 00 3 5 00 H2S, 3 comps, validation p re d ic te d
Methane content prediction H2S content prediction H2S content prediction (model trained with <2000 ppm H2S data
RMSEP: 278 ppm RMSEPCV: 294 ppm RMSEP-Test: 592 ppm R²: 0.78 R²cv: 0.75 %X explained: 94.9% %Y explained: 77.7% Saturation effect 80 0 10 0 0 1 2 00 1 40 0 16 0 0 1 8 0 0 H2S, 3 comps, validation p re dic te d H 2 S c o n ce n tra ti on ( p pm v) RMSEP: 123 ppm RMSEPCV: 132 ppm RMSEP-Test: 324 ppm R²: 0.77 R²cv: 0.74 %X explained: 94.7% %Y explained: 77.4% 0 1 2 0.0 0.5 1.0 0 10 20 30 40 50 60 70 80 90 100 110 120 N -N H 4 ( g / L ), A lk . ( m l C O 2 Time (days) 0 2 4 6 8 10 12 14 16 18 20 0 10 20 30 40 50 60 70 80 90 100 110 120 e -n o s e i n d ic a to r 1 a n d 2 Time (days) T² (indicator 1) Q² (indicator 2) limit 0 1 2 3 4 5 6 7 8 9 10 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 O L R ( g /L / d a y) , b io g a s a n d C H 4 r a te (L / L /d ay ) Time (days) OLR
Methane production rate biogas production rate
0 2 4 6 8 10 12 14 16 18 20 75 76 77 78 79 8081 82 83 84 85 86 87 88 89 90 91 92 93 94 95 9697 98 99 100 e -n o s e in d ic a to r 1 a n d 2 Time (days) T² (indicator 1) Q² (indicator 2) limit
Contact: Gilles ADAM
ULg – Arlon campus environnement 185, avenue de Longwy – B-6700 ARLON Tel. (+32) 63 230 936
Gilles.Adam@ulg.ac.be Funding statement: The research was co-financed by the European Union via the
FEDER in the framework of INTERREG IVA program.
http://www.ecobiogaz.eu/
References:
[1] Madsen, M., Holm-Nielsen, J.B., Esbensen, K.H., 2011. Monitoring of anaerobic digestion processes: A review perspective. Renew. Sustain. Energy Rev. 15, 3141–3155. doi:10.1016/j.rser.2011.04.026 [2] Adam, G., Lemaigre, S., Romain, A.-C., Nicolas, J., Delfosse, P., 2013. Evaluation of an electronic nose
for the early detection of organic overload of anaerobic digesters. Bioprocess Biosyst. Eng. 36, 23–33. doi:10.1007/s00449-012-0757-6
CONCLUSIONS
• The MOx-based e-nose appears as an interesting tool for AD process monitoring. The study showed that it was possible to quickly detect out-of-control situations (related to both biological dysfunctions and measurement system failure) using only gas phase parameters to build the model. The e-nose could be used for rapid detection of disrupting AD process. By the e-nose advice, subsequent analytical measures should be employed to determine the cause of the process disruption.
• The same MOx-based e-nose could be utilized for a rough screening of CH4and H2S concentration in the raw biogas. Even though, quality of the prediction of H2S and CH4content is low. In addition, it seems that model quality was affected
by the sensor drift. Though, quality of prediction degrades over time. Data pre-treatment should be investigated to compensate sensor drift effects.
45 50 55
measured
Phase I: disturbed (process day 78 to 98) •Alkalinity sharp drop
•Stable pH
•H2S concentration sharp increase •CO2concentration increase •Low biogas quality •High biogas production Phase II: reactor collapse (after 98 days) •alkalinity sharp drop
•pH decrease •H2S emissions
•Extremely low biogas production
0 10 20 30 40 50 1 2 3 4 number of components R M SE P train CV adjCV test measured 0 5 10 15 20 20 0 4 00 60 0 80 0 1 0 00 12 0 0 number of components R M S E P ( p pmv ) train CV adjCV test 800 1000 1200 1400 1600 1800 2000 measured H2S concentration (ppmv) CH4prediction model:
•cross-validation RMSEP is good (1.99%) whereas test set
presents a much higher RMSEP (3.7%) overfitting?
H2S prediction models:
•1stmodel trained with all H
2S concentrations of the
training data set give a RMSEP of 278 ppm whereas mean
value of H2S concentration in the training set was 1425 ppm
•2st model trained with [H2S] lower than 2000 ppm gave
RMSEPCV of 132 ppm, which is much better than the first model.
For all models:
•Test set RMSEP was much higher than cross-validation RMSEP
causes: overfitting and/or sensor drift effect
Focus on the disturbed period (days 75 to 100)
Indicators return under the control limit: process can be still recovered
Indicators do not return under control limit: collapsing reactor
Higher OLR but gas production decrease 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 10 20 30 40 50 60 70 80 90 100 110 120 H y d ro g e n s u lp h id e c o n c e n tr a ti o n (p p m v ) Time (days) measured H2S predicted H2S Training data set
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 110 120 M e th a n e c o n te n t (% v o l) Time (days) measured CH4 predicted CH4 Training data set