A Method for Monitoring the Surface Conservation of Wooden Objects

Sep 9, 2003 - attack, and to UV light) were simulated on the surface of wooden boards ... variate control charts, namely, the Shewhart, CUSUM, and. SM...
0 downloads 0 Views 194KB Size
Download PDF
Anal. Chem. 2003, 75, 5567-5574

A Method for Monitoring the Surface Conservation of Wooden Objects by Raman Spectroscopy and Multivariate Control Charts Emilio Marengo,* Elisa Robotti, Maria Cristina Liparota, and Maria Carla Gennaro

Dipartimento di Scienze e Tecnologie Avanzate, Universita` del Piemonte Orientale “A. Avogadro”, Spalto Marengo 33, 15100 Alessandria, Italy

The principles of quality control and multivariate statistical analysis were applied to the monitoring of the conservation state of wooden works of art. Three degradation processes (exposure to a wet atmosphere, to an acidic attack, and to UV light) were simulated on the surface of wooden boards of the XVI century and were monitored by the use of Raman spectroscopy. The resulting spectra were treated by principal component analysis, which allowed reduction of the system dimensionality. The relevant principal components were used to build multivariate control charts, namely, the Shewhart, CUSUM, and SMART (simultaneous scores monitoring and residual tracking) control charts. The Shewhart charts allowed identification of the effects due to the degradation processes. The CUSUM charts permitted identifcation of the exact moment in which the degradation treatment was started. The SMART charts provided a synthetic description of the conservation state of the wooden surface by means of only two charts: the T2 Hotelling and the DModX charts. Increasing attention is paid to the studies devoted to heritage goods preservation.1-10 Investigations have been performed to simulate in the laboratory the main degradation processes affecting works of art, as those caused by exposure to light, wet atmosphere, and acidic and basic attacks.1,2 Such studies have recently acquired even more importance since it has been proved * Corresponding author. Fax: (+39) 0131 287416. E-mail: [email protected]. (1) Athanassiou, A.; Hill, A. E.; Fourrier, T.; Burgio, L.; Clark, R. J. H. J. Cultural Heritage 2000, 1, S209-S213. (2) Fotakis, C.; Anglos, D.; Balas, C.; Georgiou, S.; Vainos, N. A.; Zergioti, I.; Zafiropoulos, V. OSA TOPS Laser Opt. Manuf. 1996, 9, 99-104. (3) Bacci, M.; Picollo, M.; Porcinai, S.; Radicati, B. Thermochim. Acta 2000, 365, 25-34. (4) Thomson, G. The Museum Environment, 2nd ed.; Butterworth/Heinemann: Oxford, U.K., 1986. (5) Bugio, L.; Ciomartan, D. A.; Clark, R. J. H. J. Mol. Struct. 1997, 405, 1-11. (6) Clark, R. J.; Gibbs, P. J. Anal. Chem. 1998, 70, 99A-104A. (7) Clark, R. J. H.; Curri, M. L.; Langanara, C. Spectrochim. Acta A 1997, 53, 597-603. (8) Andalo`, C.; Bicchieri, M.; Bocchini, P.; Casu, G.; Galletti, G. C.; Mando`, P. A.; Tardone, M.; Sodo, A.; Plossi Zappala`, M. Anal. Chim. Acta 2001, 429, 279-286. (9) Baronti, S.; Casini, A.; Lotti, F.; Porcinai, S. Chemom. Intell. Lab. Syst. 1997, 39, 103-114. (10) Orlando, A.; Picollo, M.; Radicati, B.; Baronti, S.; Casini, A. Appl. Spectrosc. 1995, 49 (4), 459-465. 10.1021/ac0300791 CCC: $25.00 Published on Web 09/09/2003

© 2003 American Chemical Society

that works of art can also undergo degradation processes in museums, because of exposure to unsuitable light or atmosphere.3,4 Together with methods for simulating the degradation processes, methods for the evaluation of the actual preservation state of heritage goods must be developed. Nondestructive methods must be applied to avoid the paradox of damaging a work of art while monitoring its preservation state. Nondestructive techniques, such as near-infrared (NIR) and Raman spectroscopy, have been applied to the preservation of heritage goods and, in particular, of paintings, that show high sensitivity to environmental damage.5-7 The use of these instrumental techniques provides complex data sets, constituted by the spectra of the samples. These data sets are suitable to be treated by multivariate analysis techniques such as principal component analysis (PCA).9-10 A powerful tool for the evaluation of degradation processes occurring on a work-of-art surface can consist of the application of the principles coming from statistical process control.11-13 Lending the principles of quality assessment from quality control, a degradation process can be considered as a deviation from the initial “in-control” state, characterized by the absence of damages. From this starting point, it is possible to build control charts able to represent the situation of the system before and during the degradation. The aim is to evaluate whether a relevant damage is occurring and possibly to prevent it. Given the complexity of the data set, PCA and other related techniques can also be useful for this purpose: multivariate Shewhart control charts14,15 can be constructed on the relevant principal components (PCs) obtained from the analysis of the spectral data set. Shewhart and CUSUM (cumulative sums)16 multivariate control charts can allow detection of the presence of active degradation processes and possibly identification of their causes as well. In the case of wooden works of art, the objects are usually treated at the finishing stage with some conservation agents, (11) Shewhart, W. A. Economic Control of Quality of Manufactured Product; Van Nostrand: Princeton, NJ, 1931. (12) Montgomery, D. C. Introduction to statistical quality control, 3rd ed.; Wiley: New York, 1991. (13) Ryan, T. P. Statistical Methods for Quality Improvement; Wiley: New York, 1989. (14) Kourti, T.; MacGregor, J. F. Chemom. Intell. Lab. Syst. 1995, 28, 3-21. (15) Alt, F. B.; Smith, N. D. Multivariate process control, In Handbook of Statistics; Krishnaiah, P. R., Rao, C. R., Eds.; North-Holland: Amsterdam, 1988; Vol. 7, pp 333-351. (16) Woodward, R. H.; Goldsmith, P. I. Cumulative Sum Techniques; Oliver and Boyd: London, 1964.

Analytical Chemistry, Vol. 75, No. 20, October 15, 2003 5567

directly in the artist’s workshop: this certainly influences the Raman spectra. A careful analysis of the spectra may shed light on the finishing procedures. Moreover, the application of multivariate control charts should allow us to distinguish deterioration processes affecting the raw materials of the work of art and or the finishing substances. The present study represents a preliminary approach to the study of wooden works of art. Many studies have been performed, so far, concerning the degradation of wood by Raman and NIR spectroscopy,17-20 but this is the first example of application of multivariate techniques and control charts to monitor the deterioration processes. In our laboratories, different accelerated aging processes on wooden board surfaces (i.e., exposure to wet atmosphere, to UV light, and to acidic attack) have been simulated and monitored by Raman spectroscopy. The data sets produced have been analyzed by PCA, and quality control principles have been applied to the relevant PCs. Shewhart charts and CUSUM control charts were then built for each simulated process on the basis of PCA results. The so-called SMART (simultaneous scores monitoring and residual tracking)21-23 charts were investigated as well. They consist of two control charts: the T2 Hotelling and the DmodX control charts. The SMART charts present the advantage of describing the entire process with the aid of only two control charts, despite the great data set complexity. THEORY Chemical data sets are often characterized by a large number of variables, with complex correlation patterns. In these cases, the application of a multivariate approach24-26 represents the best procedure for monitoring processes and identifying trends and behaviors. In the present case, the application of a multivariate approach is even more effective because of the great number of variables (the emissions) correlated with each other. A simplification of the problem can be obtained by means of PCA,27-30 a powerful data mining technique that provides a new set of orthogonal variables to describe the system under investiga(17) Kihara, M.; Takayama, M.; Wariishi, H.; Tanaka, H. Spectrochim. Acta A 2002, 58 (10), 2213-2221. (18) Eichhorn, S. J.; Sirichaisit, J.; Young, R. J. J. Mater. Sci. 2001, 36 (13), 3129-3135. (19) Agarwal, U. P.; Ralph, S. A. Appl. Spectrosc. 1997, 51 (11), 1648-1655. (20) Edwards, H. G. M.; Farwell, D. W.; Webster, D. Spectrochim. Acta A 1997, 53, 2383-2392. (21) Wold, S.; Albano, C.; Dunn, W. J., III; Esbensen, K.; Hellberg, S.; Johansson, E.; Lindberg, W.; Sjostrom, M. Analusis 1984, 12 (10), 477. (22) Wikstrom, C.; Albano, C.; Eriksson, L.; Friden, H.; Johansson, E.; Nordahl, A.; Rannar, S.; Sandberg, M.; Kettaneh-Wold, N.; Wold, S. Chemom. Intell. Lab. 1998, 42, 221-231. (23) Wikstrom, C.; Albano, C.; Eriksson, L.; Friden, H.; Johansson, E.; Nordahl, A.; Rannar, S.; Sandberg, M.; Kettaneh-Wold, N.; Wold, S. Chemom. Intell. Lab. 1998, 42, 233-240. (24) Pillai, K. C. S. Hotelling’s T2 Statistics. In Encyclopedia of Statistical Sciences; Kotz, S., Johnson, N. L., Eds.; Wiley: New York, 1983; Vol. 3, pp 668-673. (25) Hotelling, H. Multivariate quality control: Techniques of statistical analysis; Hastay, E., Wallis, E. D. S., Eds.; Mc-Graw Hill: New York, 1947. (26) Mason, R. L.; Tracy, N. D.; Young, J. C. J. Qual. Technol. 1995, 27 (2), 99-108. (27) Davis, J. C. Statistics and data analysis in geology; John Wiley & Sons: New York, 1986. (28) Massart, D. L.; Vanderginste, B. G. M.; Deming, S. N.; Michotte, Y.; Kaufman, L. Chemometrics: a textbook; Elsevier: Amsterdam, 1988. (29) Box, G. E. P.; Hunter, W. G.; Hunter, J. S. Statistics for Experimenters; John Wiley and Sons: New York, 1978. (30) Wold, S.; Esbensen, K.; Geladi, P. Chemom. Intell. Lab. Syst. 1987, 2, 3752.

5568

Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

tion. The new variables, called PCs, consist of linear combinations of the original variables and are built so that each successive PC explains the maximum possible amount of variance left in the original data set, after extraction of the previous PCs. The PCs represent an optimal point of view of the original data, with a substantial dimensionality reduction and redundancy elimination. It is fundamental to select only the relevant PCs, which account for the systematic information present in the complex data set. When a degradation process takes place, PCA may allow identification of the relevant changes of the surface, through the analysis of the projections (scores) of the original data (observations at different time) in the PCs space. On the other hand, the analysis of the composition of each PC in terms of the contribution of the original variables (loadings) may lead to the identification of the causes that produced the changes. Shewhart control charts can be constructed using the scores of the PCs or the residuals,31-33 the latter being the data obtained by subtracting from the original data the information contained in the set of significant PCs. The residuals account for the information not explained by the significant PCs. If a degradation process leads to the formation of new species active with respect to the analytical method, the related information shall be present in the residuals. If PCA is applied for monitoring a degradation process, the presence of groups of observations may account for “special causes” of variation, associated with the occurrence of sudden degradation effects. The indication of out-of-control processes (in this case an occurring degradation) is given by the presence of samples out of the (3σ control limits or of trends and systematic behaviors. Since multivariate Shewhart charts are constructed on the PCs, it is important to stress that the presence of out-of-control samples can be due not to the value of a single original variable but to the values of the variables whose loading is larger for the PC considered. So this method implicitly takes into account the correlation structure of the variables. Shewhart control charts are not very sensitive to small changes of the process mean. To effectively detect the occurrence of such changes in the process, a powerful tool is the so-called CUSUM control chart.16 This chart represents the cumulative deviation of each successive observation from a reference value, for instance, the average. In the case of multivariate CUSUM charts, one chart is built for every significant PC and the reference value is the average score for the corresponding PC. This chart is very sensitive with respect to slight changes of the observations average and the occurrence of a particular event is registered as a change of the steepness in the correspondent graph, which shows, in this case, a clear trend whose steepness is proportional to the change of the mean of the variable considered. Another powerful tool for analyzing multivariate processes is the so-called SMART charts for monitoring scores and residuals at the same time. The two control charts constituting the SMART charts were originally described by Wold.21-23 The first chart consists of a typical T2 Hotelling control chart, built by considering all the significant PCs. (31) MacGregor, J. F.; Kourti, T. Control Eng. Pract. 1995, 3, 403-414. (32) Hayter, A. J.; Tsui, K. L J. Qual. Technol. 1994, 26 (3), 197-208. (33) Mastrangelo, C. M.; Montgomery, D. C. Qual. Relat. Eng. Int. 1995, 11 (2), 79-89.

Table 1. Percent Explained Variance and Percent Cumulative Explained Variance for the First Six PCs Calculated for the Three Treatments Performed (a) wet atmospherea

PC 1 PC 2 PC 3 PC 4 PC 5 PC 6

(b) acidic attackb

(c) UV lightc

% explained variance

% cumulative explained variance

% explained variance

% cumulative explained variance

% explained variance

% cumulative explained variance

57.87 13.42 7.81 5.02 4.18 2.81

57.87 71.29 79.10 84.12 88.31 91.12

66.27 8.04 5.20 4.62 3.65 3.07

66.27 74.31 79.52 84.14 87.79 90.86

59.10 12.12 5.91 4.60 3.80 2.97

59.10 71.22 77.12 81.72 85.53 88.50

a Averaged spectra, 39: 18 spectra of the characterization measures, 12 of the treated measures, and 9 of the molded measures. b Averaged spectra, 84: 15 spectra of the characterization measures and 69 of the treated measures. c Averaged spectra, 48: 15 spectra of the characterization measures and 33 of the treated measures.

The second SMART chart is the DModX control chart, which represents the residuals calculated by the application of the SIMCA (soft independent model of class analogy)34,35 classification model based on the significant PCs. The residuals reported in this chart represent the distance of each sample from the SIMCA model. The SMART charts are powerful tools for process control because they collect all the significant information in only two graphs that show the process in its whole. A possible disadvantage of these charts consists of the more difficult interpretation of the final result due to the combination of the information of all the significant components together. EXPERIMENTAL SECTION The three simulated extreme environmental situations, namely, wet atmosphere, acidic attack, and UV light, were performed on three different wooden boards belonging to a beam of the XVI century. To evaluate whether some relevant effects had occurred on the sample surface it was necessary to characterize the samples (one board for each treatment) before applying any treatment. This was achieved by performing 15 measurements on each board. Each Raman analysis was repeated three times on the same position of the sample; the three spectra were then averaged and considered as a single measurement. The Raman spectra were registered from 4000 to 50 cm-1, with a resolution of 1 cm-1. This produced data matrices containing 3950 variables: i.e., the emissions measured at 3950 wavelengths. A baseline correction was applied to each spectrum in order to eliminate the large band due to the fluorescence of the sample. A smoothing procedure was applied to each spectrum in order to reduce its problem dimensionality and to simplify the statistical treatment: every 20 wavelengths, the absorbance intensities were substituted by their averaged value leading to data sets containing 200 variables (the first 7 wavelengths were cut off because they did not show any relevant band). Figure 1 represents a typical Raman spectrum for a chestnut tree sample, with the corresponding band assignments, which reveal the presence of bands belonging to both lignin and cellulose. (34) Wold, S. Kem. Kemi. 1982, 9, 401. (35) Frank, I. E.; Lanteri, S. Chemom. Intell. Lab. 1989, 5, 247.

Figure 1. Raman spectrum for a chestnut tree sample, with the corresponding band assignments.

Wet Atmosphere. The exposure to moisture and rain was simulated by exposing the wooden board to the vapors deriving from a water bath at 60 °C. Raman spectra were registered after 1, 4, 22, and 37.5 h of exposure to the wet atmosphere. The wet board was then swaddled with Parafilm and stored for 55 days, during which a yellow mold appeared on the surface. The sample was analyzed by Raman spectroscopy every 15 days. Acidic Attack. The effect deriving from the exposure to acidic rains was simulated by treating the wooden board with 1.0 × 10-4 M H2SO4. The Raman spectra were registered every 2 h of exposure to the acidic solution, for a total of 46 h. Exposure to UV Light. The simulation of the exposure to sunlight was achieved by exposing the wooden board to a UV lamp emitting at 234 nm (15 W) (Osram), registering the Raman spectra every 4 h, for a total of 44 h. Raman Spectroscopy and Software. The spectra were registered by a RFS 100 FT-Raman Michelson spectrophotometer equipped with a Nd:YAG laser at 0.350 mW, with a large emission intensity at 1.064 nm. The detector is a Ge diode kept cool by liquid nitrogen at 77 °K. The spectrophotometer is directly controlled by a PC with the software OPUS 3.1 (Bruker). For principal component analysis and control charts, the following softwares were used: Statistica version 5.1 (StaSoft Inc.), The Unscrambler version 7.6 (Camo Inc.), Microsoft Excel 2000 (Microsoft Corp.), and Matlab version 6.1 (The Mathworks). Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

5569

RESULTS AND DISCUSSION PCA was performed on each group of the characterization measurements. The relevant PCs were then used to project all the samples (both those used for the characterization and the treated ones) in the new variable space, to evaluate whether some differences due to the treatment could be detected. All data sets were autoscaled before the PCA calculation. PCA was applied for both complexity reduction and monitoring purposes; the results obtained by means of PCA, i.e., the scores representing the measurements in the new space given by the relevant PCs, were used to build multivariate Shewhart and CUSUM control charts and SMART charts. The results of PCA performed for each treatment are given in Table 1. The percent explained variance accounts for the amount of information contained in the original data set, which is explained by each PC, while the percent cumulative explained variance accounts for the amount of information present in the original data set, which is described on the whole by the first n PCs. These parameters are useful for the choice of the relevant PCs, i.e., those PCs accounting for the largest amount of information. Wet Atmosphere. The results of PCA performed on the characterization samples are given in Table 1a. The first two principal components were considered relevant for the successive statistical analysis, since they explain 71.30% of the total information contained in the data set. Figure 2a represents the scores and the loadings of the first two relevant PCs. The loadings plots are represented along each axis. From the loadings we can state that the last part of the spectrum from 2000 to 3985 cm-1 weighs more heavily on PC1, together with the regions 250-400 and 860-1750 cm-1. On PC2, the regions 350-500, 1070-1700, and 2800-3000 cm-1 show large positive loadings, while the regions 150-280, 1805-2000, 23002700, and 3300-3450 cm-1 have large negative loadings. The score plot shows three clusters of samples: the characterization samples (group A) are grouped around the origin of the axes; the observations move slightly toward larger negative values on PC2 as the exposure to wet goes on (group B), showing an increase of the emission in the regions 150-280, 1805-2000, 2300-2700, and 3300-3450 cm-1 and a decrease in the regions 350-500, 1070-1700, and 2800-3000 cm-1. The formation of the yellow mold onto the sample surface causes a shift of the scores of PC1 toward larger positive values and a shift of the scores on PC2 toward larger negative ones (group C). Acidic Attack. The results of PCA performed on these data sets are reported in Table 1b: the first two principal components explain 74.31% of the total variance and have been used for the successive study. The loadings and the scores plots are represented in Figure 2b. The analysis of the loadings of the first two components suggests that, on the first principal component, the spectral regions 300-1750, 2500-2800, and 3450-3985 cm-1 have large positive values while on the second component the regions 240300, 1750-2000, and 3200-3400 cm-1 have large positive values. The regions 130-150, 533, 1600, and 2850-3100 cm-1 have large negative values. In the score plot of PC1 and PC2, the samples are grouped in two clusters: the characterization samples are grouped around the origin of the axes, while the treated ones are progressively shifted toward larger negative values on PC1 and 5570 Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

Figure 2. Loadings plots and scores plot for the first two significant PCs for the three applied treatments: (a) exposure to the wet atmosphere (CAR, characterization samples; WET, samples exposed to the wet atmosphere; MOU, molded samples); (b) exposure to the acidic attack; (c) exposure to UV light.

larger positive values on PC2. The treatment causes the progressive decrease of the emission in all the spectral regions, except for those in the ranges 240-300, 1750-2000, and 3200-3400 cm-1, which increase during the acidic treatment.

Figure 3. Multivariate Shewhart control charts for the samples exposed to the three treatments: wet atmosphere (a); acidic attack (b); UV light (c) (dotted lines identify the beginning of the degradation process).

Exposure to UV Light. Table 1c reports the results obtained by performing PCA on these data sets; the first two components explain 71.22% of the total variance and are used for the successive discussion. The loadings and scores plots of the first two PCs are represented in Figure 2c. The first component shows large positive loadings for most of the spectral region. The second component

carries the information contained in the spectral regions 128430, 1050-1750, and 2850-3000 cm-1 (positive loadings) and 630-1000 and 2600-2800 cm-1 (negative loadings). Multivariate Control Charts. The Shewhart charts were calculated from the scores of the first two components for the three applied aging treatments. The charts are grouped in Figure 3. During the exposure to the wet atmosphere (Figure 3a), a Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

5571

Figure 4. Multivariate CUSUM control charts for the samples exposed to the three treatments: wet atmosphere (a); acidic attack (b); UV light (c) (dotted lines identify the beginning of the degradation process).

decrease of the scores is observed: in fact, during the treatment, the spectra show a decrease of emission in the region ranging from 2000 to 3985 cm-1, together with decreasing emissions in the regions 250-400 and 860-1750 cm-1. The trend is rapidly inverted when the yellow mold appears on the board surface and increase of the emission in the spectrum is accompanied by increase of the scores. The last three points appear as out of 5572

Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

control with respect to the 3σ Shewhart rule. The Shewhart chart of the scores for PC2 shows an opposite behavior: the first points related to the exposure to the wet atmosphere lay around the mean value; the mean shifts toward smaller values during the exposure to the wet atmosphere, showing a decrease of the scores caused by an increase of the absorbance in the regions 150-280, 1805-2000, 2300-2700, and 3300-3450 cm-1 and a decrease in

Figure 5. SMART control charts for the samples exposed to the three treatments: wet atmosphere (a); acidic attack (b); UV light (c) (dotted lines identify the beginning of the degradation process).

the regions 350-500, 1070-1700, and 2800-3000 cm-1. The mean stabilizes at a smaller value until the sixth measurement after the appearance of the yellow mold, and then a further sudden decrease in the scores is registered. This effect is larger as the mold increases so that the last points appear as strong out-of-control.

Analysis of the Shewhart control charts of the first two PCs referring to the acidic attack (Figure 3b) gave a similar result: in regard to the first component, the initial stable trend (during the characterization step) suddenly changes after the third treated measurement, with a consequent change of the mean, which shifts Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

5573

to smaller values. The second component shows an opposite trend, characterized by the shift of the mean value to larger values. In the case of exposure to UV light, the Shewhart control charts (Figure 3c) did not evidence any relevant change of the surface. In both cases, the trend is quite stable and no out-of-control can be detected. Similar conclusions can be obtained based on the multivariate CUSUM charts represented in Figure 4. In these charts, events occurring on the sample surface are evidenced as trends and changes in steepness. For the first two treatments applied (wet atmosphere and acidic attack), CUSUM charts of both PC1 and PC2 allowed the identification of the moment when a change on the surface occurred: this corresponds for both treatments to the first sample belonging to the applied chemical attack. The CUSUM chart of the first PC of the exposure to a wet atmosphere allowed the identification of two events: a first change of the surface due to exposure to the wet atmosphere and a second change related to the appearance of the yellow mold on the surface. For what concerns exposure to UV light, only the CUSUM chart of PC1 can help in the identification of some changes occurring on the surface: two trends (first increasing and then decreasing) can be detected during the characterization step, while a continual increasing step is registered during the whole UV treatment. No evident trend can be identified by the CUSUM chart of PC2. The SMART charts for the three performed treatments are represented in Figure 5: in all of the cases, the charts were calculated on the basis of the first two PCs. The DmodX chart was built on the SIMCA model, constructed with two PCs on the basis of only the characterization samples. Again, the T2 Hotelling and the DmodX charts confirm the previous conclusions. The T2 Hotelling charts of the first two treatments show stable behavior during the characterization step followed by a continued change during the treatment. In the case of the wet atmosphere, the values decrease continuously but not abruptly; in the case of the acidic attack, the trend suddenly changes and the values increase with the first degradation sample. The DmodX charts for these two treatments confirm the conclusions: the observations belonging to both wet and acidic treatments are not well-described by the SIMCA model built on the basis of only the characterization samples. For what concerns exposure to UV light, the T2 Hotelling chart (Figure 5c) does not show any evident trend, and even if a large variability is registered, no relevant differences due to the

5574

Analytical Chemistry, Vol. 75, No. 20, October 15, 2003

treatment can be evidenced. The DmodX chart shows instead a slight increase of the values during the treatment and some values appear out-of-control, even if the difference between characterization and treated samples is much less evident than in the previous cases. Both wet and acidic treatments produced a relevant change on the sample surface, identified by means of the multivariate Shewhart and CUSUM charts and by the SMART charts: these tools thus appeared useful and effective for the identification of the effects produced by degradation treatments applied on a sample surface. Analysis by the multivariate control charts evidenced, in addition, that no effect resulted from the exposure of the wooden samples to UV light. CONCLUSIONS Three degradation processes that may occur to a wooden work of art have been simulated in our laboratories and applied to small wooden board samples. The treatment has been monitored by Raman spectroscopy, and the resulting data have been analyzed by PCA. This multivariate data analysis technique was a successful tool for data simplification (the information carried by the original 200 variables has been condensed in only two PCs for each degradation attack studied) and for identification of the spectral regions involved in the changes due to the applied treatments. Multivariate Shewhart and CUSUM control charts allowed the identification of the events related to the attacks and permitted us to state that an appreciable effect had been recorded for all the simulated treatments, except for the exposure to UV light, which produced only slight changes on the wooden board surface. The CUSUM charts permitted identification of the sample where the treatment was started, so that they can be very useful for deciding when an intervention must be applied. Multivariate SMART charts were shown to be effective in each data set description, since they allowed us to condense the amount of information carried by the first two components with the aid of only two control charts. They allowed identification of the events related to the degradation effects on the wooden samples, confirming the conclusions arrived at by multivariate Shewhart control charts.

Received for review March 3, 2003. Accepted July 10, 2003. AC0300791