M.N. Pons, S. Le Bonté, O. Potier Laboratoire des Sciences du Génie Chimique, CNRS-ENSIC-INPL, Nancy Adaptive Principal Component Analysis for toxic event detection
Introduction New regulations: treatment in adequate facilities of all incoming waters stricter limits on effluent quality, on sludge Crisis: rainstorm accidental release of toxic components some may be forecast (fire water) other not
A short selection of potential toxics Heavy metals: Hg, Cr, Pb, Cd, Zn, Cu... Solvents: white spirit,... Pesticides Herbicides Motor fuels: diesel oil,... Detergents Dyes
Introduction New regulations: treatment in adequate facilities of all incoming waters stricter limits on effluent quality Crisis: rainstorm accidental release of toxic components some may be forecast (fire water) other not Improvement of plant control strategy New scenarios
Introduction Characterisation of wastewater composition COD, BOD 5, SS N T, NH 4 +, NO 3 - P T, PO 4 - K, Ca, Mg,... Heavy metals (Cu, Zn, Cd, Hg, Cr, …) Micropolluants Some are time-consuming Some are very specific
Introduction Global (and faster) measurements temperature, conductivity, pH, redox turbidity light absorbance fixed wavelength spectra respirometry buffer capacity ... On-line In-line (sampling)
Introduction Three methods under test Respirometry Absorbance spectra Buffer capacity Multivariate data analysis method Validation on simulation Experimental validation Conclusions
Respirometry test: experimental set-up
Respirometry test DO probe sludge + substrate Typical response curves
Characteristic parameters OUR curve 4 parameters Maximal value of Oxygen Uptake Rate Oxygen volume (VO2) (5 or 15min) Peak width Initial slope
Experimental results + CuSO 4
Experimental results + dye
2 respirometers in parallel toxics added in one respirometer CuSO 4 NaOH HCl White Spirit javel Gasoil Experimental results
UV-visible spectrometry
210 nm 220 nm 254 nm 270 nm UV-visible spectrometry Anthropogenic substances
UV-visible spectrometry 210 nm 220 nm 254 nm 270 nm Detergents
UV-visible spectrometry Dyes
UV-visible spectrometry Norm. Abs Abs. Abs
Buffer capacity Normally measured Wastewater pH Alkalinity Here Acidification (pH 3) Titration to pH 11 Buffer capacity versus pH
Buffer capacity
Fault detection background Univariate SPCMultivariateSPC Overload of data PLS Partial Least Squares Projection to Latent Structures PCA Principal Component Analysis Continuous process (steady state) Kresta et al. (1991): fluidized bed and extractive distillation column Batch and Fedbatch Lennox et al. (1999): Fermentation processes ? ? Wastewater treatment plant = continuous process but not at steady state
Adaptive PCA Diurnal cycle 1 sample / 30 min (48 samples / day) or / 1hr (24 samples / day) 4 Principal Variables (PV i ) : Our ex max, Ourex T, Slope, Width ( 15 min) In the case of 1 sample / 1 hr, the samples j to j+23 are used and 2 PCs are considered: PC 1 = 1 PV 1 + 1 PV 2 + 1 PV 3 + 1 PV 4 PC 2 = 2 PV 1 + 2 PV 2 + 2 PV 3 + 2 PV 4 At sample j+24: prediction PC 1 (j+24) = 1 PV 1 (j) + 1 PV 2 (j) + 1 PV 3 (j) + 1 PV 4 (j) PC 2 (j+24) = 2 PV 1 (j) + 2 PV 2 (j) + 2 PV 3 (j) + 2 PV 4 (j) At sample j+24: actual PC ’ 1 (j+24) = 1 PV 1 (j+24) + 1 PV 2 (j+24) + 1 PV 3 (j+24) + 1 PV 4 (j+24) PC ’ 2 (j+24) = 2 PV 1 (j+24) + 2 PV 2 (j+24) + 2 PV 3 (j+24) + 2 PV 4 (j+24)
Adaptive PCA Prediction error = Detection (Q statistic) SPE = [PC 1 (j+24) - PC ’ 1 (j+24)] 2 + [PC 2 (j+24) - PC ’ 2 (j+24)] 2 Update of i, i, i, and i using samples j+1 to j+24
Adaptive PCA CP 1 CP 2 σ 1, μ 1 h h+1 h+2 h+3 h+4. h+23 h+24 σ 2, μ 2 h+25 σ 3, μ 3 …etc......
Effect of slow change in plant state PCA on 24 previous samples (1 sample/hr), estimation of actual sample
Why simulating ? Unsteady state Many factors to examine: Location of sludge sampling Ratio sludge / raw water Quality of detection in function of the toxic conc. and nature, release time and type …. Experiments on the real plant should be carefully selected « Experiments » on a simulated plant
Plant layout Incoming water to be tested Secondary settler External recycle Aeration tank Biomass sample Primary settler Wastage flow River Biomass sample
Concentration of toxic Release profile Concentration Detection
Toxic release time Detection Release time Release profile
Toxic release time Detection = 1.49 (0.07)Detection = 2.77 (0.17)
Normal situation Normal 24hr cycle: dry weather normal activity
Normal situation 5 initial variables : OURend, OURmax/A 254, VO 2 /A 254, width et A 254
Critical situation: heavy metals HgSO 4 6 mg/l 30 mg/l K 2 Cr 2 O 7 6 mg/l
Critical situation: diesel oil Addition of various amounts of diesel oil
Critical situation: white spirit Addition of various amounts of white spirit very strong inhibition
Buffer capacity 4 initial variables : pH, β (pH=4,75),β (pH=7,21), β (pH=9,25) SPE = [PC 1 (h) - PC’ 1 (h+24)] 2 + [PC 2 (h) - PC’ 2 (h+24)] 2
Buffer capacity 5-6 Nov.2001, 14h : Wastewater + citrate
UV-visible spectrophotometry
Conclusions Global (and rapid) characterization of the composition of wastewaters Absorbance spectra - Buffer capacity - Respirometry + Classical measurements (T, pH, rH, …) + flowrate + rainfall Combined with statistical methods Community activity (design, control, critical situation) We wish to thank the Grand Nancy Council for its help GEMCEA, LCPC, NANCIE the students and colleagues
Plant model 2D models for the primary settler (Stokes) and the final clarifier (Takacs et al.) Reactors in series with backmixing = f(flowrate, aeration rate) Basic control on sludge wastage IAWQ ASM 1 + inhibition : growth rate of heterotrophs and autotrophs death rate degradation of toxic Influent description COST 624 Benchmark Functions describing the Nancy WWTP effluent Respirometer model FORTRAN code on PC