THE NEW TECHPAP NIR SPECTROSCOPY FOR RECYCLED BALE INSPECTION Speaker: Authors: Didier Rech (Techpap) Alain Cochaux (CTP France) Pascal Borel (CTP France) Guy Eymin Petot Tourtollet (CTP France) Didier Rech (Techpap) SUMMARY This paper sums up all the different steps broached in the MONITOR project, for the development of a new sensor, based on NIR spectrometry, for the direct control at delivery of recovered paper bales in term of moisture, unusable materials and grade (according to EN643). The NIR spectroscopy technique has been studied and implemented using a mobile spectrometer device and different optical materials. In add, methods, based on statistical data analysis (in particular PLS regressions), have been investigated. First trials, using these techniques and methods, have been carried out in laboratory in order to study the recovered papers quality, in terms of moisture content and sample composition (paper, board, unusable materials). Then, a first prototype of an industrial sensor has been built and tested directly in a mill. Finally, a last version of this industrial sensor has been developed, and different trials campaigns (still in progress) with it have been carried out in mills. KEYWORDS Direct control of recovered paper bales ; NIR spectroscopy ; Humidity measurement. 1. INTRODUCTION For papermaking, the use of recovered papers and boards as raw materials is increasing and according to the second declaration of the Confederation of the European Paper Industry, the recycling rate will be 66 % in 2010. As a consequence, we assist more and more at an all-out paper collection and the quality of the recovered paper bales is often very variable. In order to standardize the grades recovered according to the various collection system, it exists a European standard called EN643. Its aim is to classify precisely the types of recovered papers and boards using several criteria, essentially based on their content: sorts of papers / boards, respective ratio, origin, aspect, presence of impurities and quantity And another crucial interest for papermakers is the measurement of moisture content, because of economic reasons. Indeed, the price of a recovered paper bale depends on its EN643 classification and its
weight, taking into account the moisture percentage. And to pay water instead of raw material is not very interesting for papermakers Nowadays, the control of the recovered paper bales at delivery, in order to measure moisture, presence of contaminants and more generally for coherence with EN643 standard, is essentially done by an operator using visual and/ or manual procedures. For instance, to control the compliance with specifications and conditions (EN643 standard) and the presence of contaminants, only visual appreciations of general aspect of a bale are often used. Very seldom, bales are opened and sorted manually which is time-consuming Another example is for moisture measurement: papermakers use capacitive gauge and dielectric or perform a core sampling in bale (using a coredrilling device) in order to use then classical manual procedures to determine the moisture content (gravimetric method). In conclusion, current procedures for control of recovered papers are usually manually and visually made and therefore time-consuming, tiresome and furthermore poorly objective. So only few controls are possible and these methods have a very low efficiency. As a consequence, papermakers are very interested to own an easy-to-use device allowing direct controls at delivery, rapid, objective and reliable. In order to develop a sensor able to do all these direct controls, we have worked, for some time in the scope of the project called MONITOR, on recovered paper bales characterization using a particular technique called Near Infra- Red (NIR) spectroscopy. More precisely, the aim of this project is to determine, for each recovered paper bale at delivery, its composition in term of raw materials (paper / board), unusable materials (metals, plastics ) and to measure its moisture content. And our final purpose is to integrate all these developments into an industrial sensor based on a coring device. The objectives are at different levels. Firstly, the possibility to control efficiently the bales at delivery should allow to pay raw material at its right price. Secondly, it s possible that the anticipated knowledge of the raw material quality should allow a better optimisation of the treatments and the productivity. Finally, the use of an objective method will allow to assure better relationships between suppliers and papermakers. 2. SOME GENERALITIES ABOUT NIR SPECTROSCOPY AND DATA ANALYSES 2.1. Generalities The Near Infra-Red spectroscopy has been already used for several years in different sectors like food processing, chemical or drugs. Indeed, it is a very powerful technique to study precisely and quickly the composition of samples. In particular for moisture measurement, it is well known that NIR spectroscopy is really fine-tuned. That s the principal reason why we have decided to invest in this technology. As for papermaking sector, it is really an emerging technique for a few years, and it is definitely a very promising technology for the future. 2.2. Some principles This technique is based on the measurement of reflected light (in our case) given by a sample. First of all, the sample to study must be enlightened by a light source with a wavelength (λ) range between 800 and 2500 nm. Then, the light reflected by the sample is collected by a detector and transformed into a spectrum. It is the role of the spectrometer (Figure 1).
Figure 1. General principle of NIR spectrometry A particularity of the NIR spectrometry (compared to other spectrometries) concerns the exploitation of the spectra. Indeed, it is very difficult to use them directly and often we need data analyses methods in order to develop multilinear statistics models to predict wanted outlet data (Figure 2). Figure 2. Prediction models Finally, these methods need to do a very important step of calibration in order to obtain the coefficients of the model. To do this learning, we need to know, for some particular samples, their spectra and the outlet data corresponding. Using together the inlet and the outlet data, we can build the prediction model. Note that this calibration is certainly a very long step to develop the models, it requires a lot of time, but it is really crucial for the final result, and the good efficiency of the sensor. 2.3. Main advantages The main advantages of the NIR spectrometry are : powerful technique to study precisely samples composition, no need of sample preparation, low price, very rapid measurements: 50 ms / spectrum (ie 200 measurements possible in only 10 seconds). 3. THE TRIALS PERFORMED IN LABORATORY AND THE FIRST RESULTS OBTAINED WITH THE NIR SPECTROSCOPY First studies were done in laboratory in order to study in details the correlations between paper NIR spectra and : Moisture classical measurements
Samples compositions (raw and unusable materials). In this view, an automatic spectra acquisition device was developed. Its aim is to simulate the comportment of an industrial coring device used with a standard NIR equipment. The Figure 3 gives precisions about this laboratory simulator. Figure 3. Details of the laboratory prototype Features about this acquisition system : the pipe diameter has been chosen in order to be equivalent to an industrial coring device, the sample could be moved in the two directions (up and down) with an adjustable constant speed, the complete acquisition/treatment time for one sample is about 50 ms, the measurements have been done on the slice of the samples, in order to avoid the influence of the area state, all this system is completely automatic. Using this prototype, we have prepared a test plan using this NIRS approach. Here are the results obtained.
3.1. Humidity measurement For the learning procedure of a humidity prediction model, we need samples with different humidity. In this view, we have made paper slices with standard paper for printer or photocopier, and we have used a water manual spray to obtain different humidity percentages. Then, for each 7 cores obtained, we have done 3 acquisitions of NIR spectra using our lab prototype (Figure 4), and an accurate traditional measurement of humidity (weighing, complete oven drying, re-weighing), in order to associate, to each spectrum, an output value to predict. Figure 4. Spectra acquisition The average spectra acquired (for each humidity value) and the corresponding humidity percentages measured are on Figure 5. Note that, for each core, the 3 acquisitions done were almost similar, and the measurements perfectly repeatable. Figure 5. Spectra of humidity
The spectra shown in Figure 5 are in accordance with literature. Indeed, we can see precisely two characteristic wavelength bands at 1440 and 1940 nm which correspond to water absorption bands. Furthermore, the different spectra obtained are classified in the right order in this two bands according to their humidity values measured. Using these data and some statistical algorithms (based on PLS1 regression), a first prediction model for humidity was developed and tested with success, because the accuracy obtained was around ± 0.35 %, which is comparable to manual measurements performed by an operator. 3.2. Unusable materials measurement As for moisture, first trials were performed in laboratory to study, more in details, spectra of different samples corresponding to different compounds. After some tests; the compounds retained were : paper/board, aluminum (metals), and three sorts of plastics (types: mugs, paper-bags, packaging). As a consequence, we have made cores with 100% of each of these compounds (for calibration), and cores with blends of several compounds (for tests). Then, using our lab simulator, we have done 3 NIR acquisitions of each core, in order to control the good repeatability of the measurements. The average spectra obtained for calibration cores are on Figure 6. Note that, for each core (not only for calibration cores but also for test cores), the 3 acquisitions performed were almost similar, and the measurements perfectly repeatable. Figure 6. Spectra of various compounds Studying more in details the spectra shown in Figure 6, we can notice that: Firstly they are in accordance with literature (in particular: typical absorption band for plastics around 1700 nm); Secondly, they are different enough in order to be distinguish automatically by a system. Finally, using these data acquired in lab and some statistical methods (based on PLS2 algorithm), a first prediction model for compounds was developed and tested with success. Indeed, it is able to distinguish with a very good precision these different compounds in a random mixing core.
3.3. Raw materials measurement (according to EN643 standard) Using EN643 standard, the distinction which is the most interesting for papermakers is certainly the distinction between 1.04 and 1.05 grades. In other words, we would like to study the possibility of using NIR spectroscopy in order to distinguish corrugated board versus all other papers. As a consequence, like for unusable materials studies, we have first done different spectra acquisitions of paper and corrugated board. The mean spectra obtained are shown on Figure 7. Figure 7. Spectra of paper and corrugated board Figure 7 shows clearly that we obtain the same curves appearances but with a little attenuation for the board which limits for a good modeling? The learning procedure to build the prediction model has been done using different spectra of different papers and corrugated boards. With statistical methods based on PLS2 algorithm (and some other specific treatments), a prediction model for these different compounds percentages has been developed. Then we have tested this prediction model using the same spectra as for learning, and other spectra acquisitions. As a consequence, the percentages predicted are quite good with paper and board used for learning, but unfortunately, the results are not satisfying with other paper compounds like newspapers, magazines, etc Indeed, the differences between paper and corrugated board in our spectra acquisitions and models are certainly more due to a difference of brightness than a difference of typical signature in the NIR area In other words, it is possible to distinguish a brown (for instance) corrugated board and a white paper, but our NIR system has a lot of problems to classify a white corrugated board in the corrugated board family, or a grey paper (like a newspaper with a lot of ink) in the paper family. 3.4. Conclusion of these first lab trials The results obtained in laboratory with the NIR technology and our automatic lab simulator have allowed us to check the use and the discriminant capability of this approach for humidity and unusable materials that we would like to characterize (vs. paper raw materials). As a consequence, these first results are really very encouraging. However, concerning the distinction between different raw materials, and in particular corrugated boards vs. all other papers, the first lab results are not very satisfying. Indeed, our systems and models developed seem to be more sensible to a difference of brightness than a real difference of raw materials. Moreover, the different methods tested in this part are very powerful, but also extremely sensitive to the measurement environment. Indeed, optic (and mechanic) for spectra acquisition, but also statistic models used, have
to be really fine-tuned in order to optimize the discrimination. As a consequence, we still had a lot of works to do for the next (and last ) step of this project : to integrate all these developments into a complete industrial device to obtain an automatic, rapid, objective and reliable sensor. 4. OUR FIRST INDUSTRIAL SENSOR PROTOTYPE AND THE FIRST TRIALS PERFORMED IN A MILL After our trials campaigns in laboratory, we have decided to develop an industrial prototype of our sensor, in order to perform our first trials in industrial condition directly in a mill. The Figure 8 shows a synopsis of this prototype. To build it, we have simply used the same optical acquisition system as for our lab simulator, fixing it at the end of the drill of an industrial coring device, and consolidating its mechanical resistance. Furthermore, a little system of cleaning was fitted to the coring device in order to clean and to protect the measurement head. All the rest of the system was sent to a few meters from the coring device, and was protected too in an industrial rack. Finally, we have implemented and used the same acquisition software as in our lab simulator in order to perform completely automatic and continuous measurements. Figure 8. Synopsis of the first industrial sensor prototype Using this prototype, trials have been carried out in a mill during one week to perform direct control of recovered paper bales at delivery. In this way, we have collected a lot of data from a lot of bales (essentially 1.04 and 1.05 sorts). Indeed, each sample (ie core) correspond to about 200 spectra, and we have acquired about 250 samples. Furthermore, for each of this 250 samples, we have determined precisely (using manual standard measurements) the output value of our prediction system (for calibration and tests), that is to say: the moisture content and the quantity of each unusable materials. Then, we have developed and tested different data processing algorithms in order to : detect automatically the beginning and the end of each core in the drill (and so the spectra to take into account for the statistical predictions), realize a first preprocessing of these spectra to obtain exploitable data, realize a second preprocessing of the spectra to select the wavelength bands to take into account in order to reduce the noise and to increase the robustness of the prediction models. With these preprocessing tools, we have noticed that the process data obtained were curves like those obtained with the lab simulator. As a consequence, we were able to apply the different models developed in lab, with only a few adaptations of them.
First of all, for moisture measurement, the predictions done were quiet good, because, for most of the case, the mean of the absolute values of the prediction errors obtained may stay around 0.5 %. Otherwise, for the unusable materials measurement, we have checked that our system is perfectly able to make a good distinction between the different compounds that we want to characterize, and in particular the cellulose materials, metals and plastics. However, different other problems have came up with this first industrial prototype : problems of clogging, problems of mechanical resistance in industrial conditions during a long time, and, above all, it was totally impossible to automate completely this measurement head in order to obtain a final and definitive complete industrial sensor. As a consequence, we have worked one more time on the optical and mechanical part of our sensor in order to develop a new version of this industrial sensor, able to solve all these troubles. 5. THE LAST INDUSTRIAL RECOVERED PAPER BALES QUALITY CONTROL SENSOR DEVELOPED The last version of our recovered paper bales quality control sensor was developed at the beginning of year 2006. A lot of important and crucial improvements were done on this prototype, in particular to increase as best as possible its mechanical resistance in industrial conditions, and to obtain a definitive and completely automatic version of this measurement head. Furthermore, this new device obtained is a little bit less expensive than the precedent one! To be more precise, two industrial solutions were developed. The first one was build to equipped directly an industrial coring-device in order to perform direct control at delivery on trucks. The second one is overall less efficient, but transportable. Figure 9 shows the synopsis of our complete sensor in the embedded version. As we can see on this draw, there are essentially two different parts. The first one is a new optical measurement head which has been adapted directly on the output pipe of a coring device. The second one is a strong industrial rack which contains all the intelligence of the device: the spectrometer, the PC CPU (for spectrometer control and data processing), and the MMI to watch, process, record, the results. Figure 9. Synopsis our last industrial sensor in the embedded version On Figure 10, we can see a picture of this complete sensor (on the left: the coring device including the measurement head; on the right: the industrial rack), and Figure 11 is a picture more precise about the complete measurement head fixed on the output pipe of the coring-device using a specific adaptation device.
Figure 10. the MONITOR sensor Figure 11. The measurement head fixed on the output pipe of the coring device Finally, the Figure 12 gives some precisions about the measurement head. As we can see on this picture, the NIR light source and the reflected signal collection are now included together inside the measurement head, and this complete device is connected to the NIR spectrometer and the power supply through a rigid 5 meters long pipe. Furthermore, the head is very light and very robust, its total length is only 20 cm, and there is also an optical device inside the head in order to optimize the light illumination and the signal collection in the optical fiber to the spectrometer. Figure 12. Some details about the measurement head Concerning the second version of this industrial sensor, it is transportable and completely manual: all the acquisition and data processing device is the same as that for the embedded version, except for the measurement head (Figure 12) which is longer (about 1 meter) in order to be introduce in a hole done in a paper bale by any perforating device. Note that this second version of our sensor is certainly a little bit less precise and efficient that the first one, but it is also much less expensive because it doesn t need an industrial coring device to use it.
6. THE INDUSTRIAL TRIALS PERFORMED IN A MILL USING THESE NEW SENSORS: FIRST RESULTS Trials have been carried out during several days in an important mill in Spain to perform direct control of recovered paper bales at delivery. In particular, our last trials campaign has allowed us to collect about 150 cores from about 70 bales (essentially 1.01, 1.02, 1.04, 1.05 and 4.02 sorts), and each of these cores correspond to about 200 spectra. Furthermore, for each of this 150 samples, we have determined precisely (using manual standard measurements) the output values of our prediction system (for calibration and tests), that is to say : the humidity percentage, the quantity of each unusable materials (metals, plastics, rests), the quantity of corrugated board and the quantity of paper (but also the classification of the sample according to EN643 standard). Here come the results obtained. 6.1. Results obtained for moisture prediction To build and to test, in industrial conditions, our prediction models for humidity, we have done some important improvement from algorithms developed previously. Then, the new models obtained have been applied in our new sensors for the mill trials campaigns. So, a lot of recovered paper bales have been analyzed in real time, and the results obtained are still very satisfying as we can see on Figure 13. The precision obtained, in not really right industrial conditions, is around ± 0.7 % which is quite satisfying, taking into account the uncertainties due to the methods, manual and automatic. Figure 13. Moisture correlations 6.2. Results obtained for Unusable materials As for moisture prediction, we have done important improvements with our data processing models (for metals and plastics) developed previously in order to fit them as good as possible to industrial conditions. And the results obtained are, one more time, very satisfying. For instance, for plastics prediction, the results obtained (Figure 14), are similar to what we can measure manually. Indeed, the precision of the prediction is around ± 0.5 %.
Figure 14. Plastic correlations 6.3. Results obtained for EN643 grades prediction To finish, concerning the distinction between corrugated board vs all other papers, and the classification according to EN643 standard, we don t think that the data acquired in the mill will allow us to obtain better results than in lab (to distinguish papers and boards). However, we have thought to study a new way to distinguish directly the different recovered paper sorts (as 1.04 and 1.05 for instance). Indeed, without knowing precisely the ratio between corrugated board and paper, it could be very interesting for papermakers to have, even so, an information about the global sort classification. This new approach is based on another statistical data treatments methods than PLS regressions. The global idea is to plot graphically, in a multi-dimensional space, the samples as particular points, and then to identify, in this space, different areas linked, each one, to a EN643 sort. Figure 15 shows such a graphical plot, in two-dimensional (X-Y) in order to make simpler the reading.
Figure 15. Two-dimensional plots of the 70 bales acquired As we can see on this graphic, unfortunately, it seems to be very difficult to identify with a good precision the different sorts. Indeed, the group points obtained for each of these sorts (for instance: blue for 1.05, or red for 1.04) don t clearly define areas enough separated. As a consequence, for the points, in particular, which are on the borderline of different areas, the error prediction will be certainly very important and not satisfying. Otherwise, increasing the dimensions number used for the modeling, we are able to obtain a better separation between the different bales sorts, but the results stay not really very good, and we reach quite quickly mathematical limits. To conclude, this new approach is at this stage not completely satisfying for us (as we can see on Figure 15). As a next step we are looking into improving the mathematical models especially the pre-processing algorithms used, in order to improve our prediction models accordingly. 6.4. Conclusions of these mill trials campaigns The results obtained during these mill trials campaigns were really very encouraging. Indeed, the complete system developed (hardware for acquisitions & software for data treatment) seems to be very efficient for humidity and unusable materials measurements. As a consequence, we have decided to fit this prototype to an autonomous use in mill by operators, in order to be able to propose this new sensor to our customers for long duration trials in industrial conditions, to check its reliability. To do that, the most important adaptation done was performed on the software interface in order to propose a sensor very easy to use, in particular for the measurement results collection. The Figure 16 shows the main window of this new software. To use the sensor, the operator has only to put the reference set (to identify the bales tested) in the white case on the left, and to click on the Start button. Then, he must perform his sampling as usual, and when he has finished, he must click on the Stop button. At this moment, the sensor performs its prediction calculations (~ 2 / 3 seconds) and, finally, plots the results obtained in the grayed cases on the right. Furthermore, the results obtained are saved into a simple text file, with the corresponding reference set, the date and the time of the measurement. This result file could be open alone, or using excel, or using any database, without any difficulties.
Figure 16. Main window of the software interface 7. LAST INDUSTRIAL TRIALS (STILL IN PROGRESS) New mechanical adaptations were necessaries to implement our new sensor prototype in this mill, in difficult industrial conditions. In particular, modifications were done on the fixing device of our measurement head, in order to build a stronger one, and we have also included directly inside this device, a specific cleaning part to avoid clogging problems. Figure 17 shows an overall picture of the complete new and last version of the Monitor measurement head, directly set up on the output pipe of an industrial coring device. On this picture, we can clearly see, among others, the new automatic mechanism of the head, associated with the specific cleaning part. Figure 17. The last version of Monitor measurement head
With this last version of our sensor, the first tests done in the mill were concluding. Indeed, we have been able to build a good calibration model for humidity, and the first tests performed with it seem to be quite good too. For instance, Figure 18 and Figure 19 show the first results obtained for moisture and plastics predictions. Figure 18. Moisture correlations Figure 19. Plastic correlations As we can see on these two last Figures, the results obtained seem to be good because we have the same precision levels as for our first industrial campaigns : around ± 1 % for humidity, and around ± 0.5 % for plastics. Since this period, the Monitor sensor is working on line, and the results obtained seem to confirm the good working order of it, in spite of different other difficulties which have appeared, have been well identified and have been solved little by little... As a consequence, we are now in a result acquisition step, and we are waiting for the final results of these trials, still in progress. 8. CONCLUSION AND PERSPECTIVES Our objective in this important project is to develop an industrial sensor able to do direct controls on paper bales at delivery, in order to determine : humidity percentage, contaminants presence and ratio, and raw materials compositions according to EN643. To achieve this goal a study has been carried out first of all at laboratory level to check the use and the discriminant capability of the NIR spectroscopy approach for humidity and unusable materials measurements. As a consequence, we have decided to instrument an industrial coring device using all these laboratory developments (hardware and software), in order to develop a first industrial sensor prototype. The results obtained with it were, again, very satisfying. Then, a new version of this industrial sensor was developed at the beginning of 2006 in order to obtain a robust and completely automatic measurement device. This last sensor has been used and tested with success, in the same mill in Spain than for our first industrial campaign. Finally, we have done some modifications on the software and the mechanical part of our sensor, in order to propose it to one French mill, for a long time trial. This trial is currently still in progress, and after having identified and corrected different technical problems, the first results obtained (based on the comparison between the results given by statistics models of our sensor, and the results manually made by an operator ) seem to be quite good too.
Finally, the last idea will be to use together the results given by this project and the results obtained in other projects (image processing in visible spectrum, volatile contaminants analysis using mass spectrometry ) in order to obtain a real identity card as precise as possible of each recovered paper bale : ratio of papers, newspapers, magazines, boards, flats, corrugated, brown, grey, contaminants, water * * * * * * * * * * The authors would like to thank the European Commission (EcoTarget project) and all the CTP Members and Partners who, involved in this project, have allowed us to do this research work, and in particular the two mills in which the different industrial trials campaigns were performed.
References: 1. Borel P., Eymin Petot Tourtollet G., Cochaux A., "instrumenter un système de prélèvement par carottage et mettre au point des procédures de contrôle de la matière première en adaptant ou développant différents capteurs aux exigences d utilisation papetière (ADEME - rapport intermédiaire)", Document CTP n 2190 juin 2005 2. Borel P., A. Cochaux, Eymin Petot Tourtollet G., " Instrumentation d'un carotteur par des capteurs permettant de contrôler les balles de papiers de récupération (ADEME - rapport intermédiaire)", CR N 4607 juin 2006 3. Borel P., "D2.1.1b Use of the Near Infra-Red Spectroscopy to qualify the grade of raw materials in recovered paper bales Technical Report (experimental results)", January 2006 (délivrable EcoTarget) CR N 4611 4. Borel P., " D2.1.2b Use of the Near Infra-Red Spectroscopy to qualify the grade of raw materials in recovered paper bales Demonstration Report (prototype deliverable)", January 2006 (délivrable EcoTarget) CR N 4609 5. Borel P., " D2.1.3b Use of the Near Infra-Red Spectroscopy to qualify the grade of raw materials in recovered paper bales Mill Trials Report)", September 2006 (délivrable EcoTarget) 6. Borel P., Eymin Petot Tourtollet, G., Cochaux A., " D2.1.4b Brown quality grade: The NIR spectrometry, a high-performance technique to qualify the grade of recovered papers and boards Presentation to Symposium", April 2006 (délivrable EcoTarget) CR N 4610 7. Borel P., Eymin Petot Tourtollet, G., Cochaux A., "CTP Monitoring strategy of recovered papers and boards: the NIR spectrometry, a high-performance technique to quantify moisture content and unusable materials", 4 th CTP/PTS Packaging Paper & Board Recycling Symposium, Grenoble, 21-22-23 March 2006 Doc CTP N 2242 8. Borel P., Sabater J., Cochaux A., Eymin Petot Tourtollet G., Veiga J., "Using NIR spectroscopy for direct control of recovered papers", 7 th CTP Recycled Fibres Forum, Grenoble, February 1 st 2005 9. Borel P., Cochaux A., Eymin Petot Tourtollet G., "The NIR spectrometry: a high performance technique to quantify moisture content, unusable materials and qualification of grade", 8 th CTP Recycled Fibres Forum, Grenoble, January 31 st 2006 10. Borel P., Cochaux A., Eymin Petot Tourtollet G., "Tomador de muestras vertical equipado con sensores infrarrojos para el control directo de la calidad de las balas de papel recuperado", 2 nd jornada técnica de control de calidad del papel recuperado Encuentro REPACAR/ASPAPEL, 15 de junio de 2006, Madrid, Espagne (congrès) 11. Borel P., Cochaux A., Eymin Petot Tourtollet G., "New quality control sensor for recycled paper bales : first industrial results, 59 ème Congrès ATIP, Bordeaux, 17 Octobre 2006 12. Borel P., Cochaux A., Eymin Petot Tourtollet G., "MONITOR : la solution embarquée pour le contrôle en temps réel de la qualité des balles de papiers récupérés", Dossier de candidature (nominé) aux Trophées ATIP 2006 de l Innovation des Technologies Papetières, Bordeaux, 18 Octobre 2006