myleaf: a predictive maintenance tool for the performance optimization of a CEM system.

Transcription

1 myleaf: a predictive maintenance tool for the performance optimization of a CEM system. L.Fiorani, S. Obaid Loccioni Group Angeli di Rosora, via Fiume 16, Ancona, Italy l.fiorani@loccioni.com Abstract Remote data control of emission monitoring systems represents an important goal in order to perform predictive maintenance actions or prompt interventions in case of system failure. In this paper are shown experimental results from real application of a multivariate analysis performed on data collected by a Continuous Emission Monitoring System (CEMS) that Loccioni has engineered and installed for a cement industry in Italy. The core of the CEMS is GIGAS 10M FT-IR Spectrometer (TÜV certified) aimed to measure HCl, HF, NO x, SO 2, CO and CO 2, equipped with a FID (Flame ionization detector) and ZrO 2 analyzer for TOC (Total Organic Carbon) and O 2 measurements respectively. All data have been collected for a period of 12 months through the Meter Data Management System (MDMS) myleaf, developed by Loccioni: 29 parameters were considered in total, including stack and cabinet sensors. Analysis was performed using principal components analysis (PCA), in order to identify those parameters that during the time period analyzed - have the greatest impact on the CEM system performances. Among these, the most important variables were identified in some FT-IR analyzer parameters, as interferometer peak and laser signal, and in the sample conditions, such as flue gas temperature. For each of these parameters it was then performed a univariate analysis, whereby it was possible to monitor their signals over time: on the basis of the collected data has been developed an algorithm able to extrapolate the trendline into the future in order to predict when there will be a possible system failure, so that it is possible to intervene in a targeted and timely manner before failure occurs. Introduction Remote control of diagnostics parameters from emission monitoring systems represents an important tool in order to perform predictive maintenance actions or prompt interventions in case of

2 system failure. The need for such request has clearly emerged by a survey conducted by Loccioni Group involving several customers. The information resulting from the inquiry have shown how a failure of the continuous emission monitoring system (CEMS) is particularly critical for the continuous production plants, especially in terms of: Loss of revenue; Personnel expenses; Cost connected with the plant restart after the failure. In order to avoid these problems, plants are usually equipped with a backup system (a system identical to the main CEMs that is put on use in case of main CEMs failure), although it means to double the cost of the emissions monitoring system. In any case, the design and develop of integrated high technological solutions for processes and environmental monitoring are not enough to guarantee their best optimal performances along time. It is important to monitor the intervention and the performances of the systems installed, in order to gather punctual information on their functionality and to perform predictive maintenance actions or prompt interventions in case of system failure. From this commitment, Loccioni developed myleaf (Measuring Your Life Energy and Future), the platform to gather and elaborate data, transforming them in meaningful information and actions. In this paper are shown experimental results from a real application of both multi and univariate analysis performed on data collected through myleaf platform on a Continuous Emission Monitoring System (CEMS). Materials and Methods The study has been performed on data acquired from a Continuous Emission Monitoring System produced by Loccioni and installed on a cement industry in Italy. The plant is aimed to produce clinker, that is a material obtained by cooking raw materials at very high temperatures (about 1250 C). The production line involves a rotatory kiln (a pyroprocessing device used to raise materials to a high temperature in a continuous process) able to produce around ton/year of clinker. The process leads to the production of exhaust gases that need to be measured, and for this purpose a continuous emission monitoring system has been installed. The core of the CEMS is LOCCIONI GIGAS 10M FT-IR Spectrometer (Fig.1a), TÜV certified, aimed to measure HCl, HF, NO x, SO 2, CO and CO 2, equipped with a FID (Flame Ionization Detector) and ZrO 2 analyzer for TOC (Total

3 Organic Carbon) and O 2 measurements respectively. The complete CEM system is represented by Loccioni GCS (GIGAS FT-IR CEM SYSTEM), as shown in the Fig. 1b. Figure 1a : Loccioni GIGAS FT-IR 10 M Figure 1b: Loccioni GCS (GIGAS FT-IR 10M CEM SYSTEM) Data related to the functionality of the analyzers have been obtained through the meter data management system (MDMS) myleaf, installed on plant in October 2012, while elaboration and data processing was conducted over a time period of 12 months (October October 2013). Below, a simple structure of the MDMS myleaf is shown : Figure 2 : Schematic representation of the meter data management system (MDMS) myleaf. A framework (that is, a data acquisition driver that communicates with the analyzers through a communications protocol) is installed on the pc of the CEMS: it acquires data concerning the functionality of the instruments, sending them to the Loccioni server. In turn, the MDMS myleaf extracts data, making them accessible through the myleaf Web Portal that is the access to the

4 interface of data collected on field, sent in real time from one or more monitored systems and store in Loccioni s database in order to be always available. The main features of the web portal are: - Data collection, filing and storage: data elaboration and graphic representation; - Visualization and evaluation of the systems performances (e.g. see Fig.3); - Interactivity: simple interaction with the system, remote access; - Data security: collected data are treated following the ruling normative. Figure 1: An example of data visualization allowed by myleaf. In this study, 29 parameters in total were considered, including stack and cabinet sensors (see Fig. 4 and Tab. 1), with a temporal resolution of acquired data of 15 minutes. More details about the most important parameters taken in account for the analysis have been reported in the Table 2.

5 Figure 2: A screen shot showing all the parameters that have been considered for the multivariate analysis.

6 Cabinet Parameters GIGAS FT-IR Analyzer Oxygen Analyzer FID Analyzer Stack Parameters Heated Line Alarm FT-IR Analyzer Sampling Line Alarm Gas Cooler Temperature Alarm Condenser Sensor Alarm Interferometer Peak Reference Peak Laser Current Laser Signal - R Laser Signal - X Laser Signal - Y IR Source Current IR Source Voltage Measuring Cell Temperature OBC Board Temperature Cell Pressure (Alarm for High pressure) Cell Pressure (Alarm for Low pressure) Measuring Cell High Temperature Alarm Lack of Flow Maintenance Maintenance Error Sampling Probe Alarm Heated Line Alarm Status of the Plant (running or stopped) Differential Pressure Absolute Pressure Flue Gas Temperature Table 1: List of the parameters. Parameter Interferometer Peak Reference Peak Laser Signal - R Laser Signal - X Laser Signal - Y Measuring Cell Temperature OBC Board Temperature Sampling Probe Alarm Heated Line Alarm Status of the Plant Description Value (in Volt) of the Interferogram Signal Value (in cm -1 ) related to the position of the interferogram Peak Three photo-diodes (R-X-Y) supplying three different analogue signals of the laser ray. These three signals are used by the OBC Board to have information about the movement of the mobile mirror, to check the circuit of dynamic alignment and to have a feedback on the quality of the signal coming from the laser Temperature inside the measuring cell (generally 180 C) The OBC (On Board Controller) Board monitors and controls optical bench and the interferometer functions. This electronics must be constantly cooled in order to avoid the overheating Alarm related to a failure of the sampling probe Alarm related to a failure of the heated line Values of 1 or 0 if the plant is running or stopped, respectively Table 2 : Brief description of the most important parameters considered for this study.

7 Because of the huge amount of data (more than one year of values collected for each of 29 parameters), the first part of the work involved a multivariate analysis in order to identify those parameters that during the time period analyzed - had the greatest impact on the CEM system performances. In order to find some hidden patterns or simplified dynamics in our data set, Principal Component Analysis (PCA) was applied: this kind of analysis was performed through SIMCA software. PCA is a mathematical procedure that transforms the correlated original variables into a generally smaller number of uncorrelated artificial variables called principal components (PC). The number of PC is less or equal to the number of original variables. The first PC captures most of total information, and each further component as much of the remaining as possible [1]. Afterwards, a univariate analysis was conducted on those parameters that have been identified as more significant by the previous analysis, in order to understand their behavior and correlations between them. Results Multivariate Analysis The graphical representation of original variables as points in the space of first two PCs (Fig. 5) can help to recognize immediately the importance of single measured FTIR parameters and a possible pattern between them. The closer a FTIR variable is to the center of the plot, the less important it is for the first two PCs, and the closer original variables are to each other, the similar they are: in this case, the behavior of one variable can affect the other one and vice versa.

8 Figure 3: Distribution of CEMS diagnostics variables within the space of the first two PC. The more the variables are far from the center the more they are influent. In addition, a further analysis involving the disposition of the density values of the variables previously considered has been performed (Fig. 6): this kind of analysis is made by transforming the values assumed by all the 29 parameters in a given instant of acquisition (every 15 minutes) in a bi-dimensional value. Every value is plotted by a point (colored according to density), and each of them represents the set of values acquired by the parameters for that instant of acquisition. The ellipse in the center of the graph represents the degree of variability (with a confidence interval of 95% 1 ), and the presence of several points outside the ellipse means likely a system failure. During the period of analysis no significant failures or breakdowns of the system occurred, indeed almost all the parameters are placed inside the ellipse. 1 The confidence limit defines the ellipse drawn around the central point that represents the normal functioning of the system.

9 Figure 4: Distribution of samples within the space of the first two PC, colored according samples density. The overall results of the multivariate analysis show how the CEMS performance are strictly linked with the FT-IR parameters and in the sample conditions, such as flue gas temperature; furthermore, no failures or breakdowns of the system occurred during the period of analysis. Univariate Analysis Afterwards, a univariate analysis involving that parameters previously identified as most important for the CEMS has been performed. One of the most important parameters is the interferometer s peak, that represents the instrumental response: this should not fall below a certain threshold (usually set at 1 V), otherwise measurements could be unstable and not reliable. The following figure shows how the interferometer peak values drop more than half a volt in less than two months, due to the drift of the signal.

10 Figure 7: Plot showing the progressive decreasing of the interferometer peak in two months. In addition to the CEMS performance performancess monitoring, which is important in order to have continuously updated detailed information about the status of the system, the analysis allows to make other interesting considerations. The following figure shows the interferogram peak s behavior from 1 July to 31 October. Figure 5: Plot of interferometer peak data for the last three months of the monitoring. After the 4th of September value has fallen down from 4.5 V to 3 V, keeping falling down continuously continuously.

11 The first thing that stands out by data processing is the different behavior of the peak s signal occurred between the 4 th and the5 th of September. Apparently this could be ascribed to some problem of the FT-IR analyzer, however by a more detailed analysis performed through the comparison with other parameters has been possible to get the correct interpretation of the observed data. More specifically, a strong correlation arises by overlapping the interferogram peak s with the flue gas temperature of the chimney stack (Fig. 9). Figure 6: Overlapping data related to the interferometer s peak (green line) and flue gas temperature (red line). In addition, the status of the plant is also showed (violet line) The figure shows that until 2 September the flue gas temperature is low because the plant was stopped (also, this is confirmed by the parameter plant status which assumed a value of zero, meaning the plant is off). After the plant restarted, the flue gas temperature increase because of the rotary kiln ignition for the clinker production: this leads to the production of gases which are measured by FT-IR analyzer, causing the drop of the interferometer peak according to the Lambert- Beer s Law [2]. Furthermore, the Fig. 9 shows a progressive decrease of the signal immediately after the restarting of the plant. On the basis of the collect data a linear decreasing has been figured out (Fig. 10): this leads to the development of an algorithm able to extrapolate the trendline into the future in order to predict when the peak will falls below the alarm threshold (typically set to 1 V).

12 FT-IR Analyzer Interferogram Peak y = x R² = /8 3/10 22/11 11/1 2/3 21/4 10/6 Figure 7: Trend of worsening of interferometer peak during the time and prevision of the time when alarm threshold will be reached Such a linear decreasing has been observed for the laser signals also (Fig. 11.), although in this case an alarm threshold is not provided: in any case this kind of analysis is useful because it allows us to have always update information about the operating status of the interferometer /6-2 5/7 25/7 14/8 3/9 23/9 13/10 2/11 FT-IR (Laser R:1217) FT-IR (Laser X:1218) FT-IR (Laser Y:1219) Figure 8: Trend of the laser signals. There are not lower alarm limits for these values, however it is possible to see the continuous worsening. In addition, another capability of such a kind of approach is to get information about the quality of maintenance performed by technicians. The following figures show two maintenance interventions, occurred between the 16 th and 18 th of October (Fig. 12) and the 25 th -26 th of September (Fig.13) respectively.

13 Figure 12: Behavior of the interferometer peak before and after the maintenance. Figure 13: Behavior of the laser signals before and after the maintenance.

14 In both cases a different trend of the signals before and after the maintenance are shown. After the maintenance of October 16 th -18 th the interferogram peak has clearly improved, by changing from a value of 1.4 V to a value of around 2.6 V. Likewise, even for the maintenance of September 25 th - 26 th an evaluation of the quality intervention is possible: in this case the alignment of the three laser signals (X-Y-R) has been obtained, in fact after maintenance the values are almost overlapped. Another useful parameter that have been considered for this data analysis is the position of the interferogram s peak, because its changes could be linked to the beamsplitter degradation. This diagnosis should be confirmed by the analysis of laser signals, that tend to decrease if the beampslitter is foggy. Throughout the period of analysis the degradation of the beamplitter did not occur (the analyzer is continuously purged with desiccated air in order to prevent the formation of moisture on the KBr surface), indeed the peak position did not vary at all, and laser signals showed no decreasing their voltage values. Figure 14: Plot showing the interferogram perk position (expressed as cm -1 ) Another correlation emerged from data analysis involves the pressure within the measuring cell of the FTIR analyzer and the OBC electronic board temperature, (that could be considered as a thermometer for the measurement of the temperature within the analyzer). Indeed, the temperature of OBC board and the other electronic components is kept low through the instrumental air:

15 sometime, however, a failure of the purging system or a decreasing of the flow can occur, leading to an increase of OBC board temperature and, thus, of the ttemperature within hin the analyzer. The behavior of such two parameters is show below (Fig. 115). Figure 15: Plot showing contemporarily both the OBC Board temperature and the measuring cell s pressure The obvious correlation tion between the two variables is due to the fact that the pressure sensor placed inside the analyzer is very sensitive to the temperature, so that a changing in the temperature leads to a wrong measurement of the pressure. This is an important evidence be because cause the quantification of the analyzed gases is carried out after normalization on pressure (and temperature). In addition, the measure of the OBC board temperature represents an indirect measure related to the status of the purging system: an increasing of the temperature it s strictly linked with a decreasing of the air flow and can be used as warning alarm before the breakdown of the purging air system occurs. At last, last useful information has been obtained by analyzing also those parameters showed no variability v during the period of monitoring. The typical example is the laser current (Fig. 16), 16) although such a behavior has been noticed even for the IR source voltage and current. Since their values have not changed at all it follows that such param parameters are not adequate for a predictive analysis of the performance of the system, even if they are useful to get information on the overall status of the emission monitoring system.

16 Figure 16: Values of laser current. No significant change during the period of analysis is observed. Conclusions In this paper experimental results of data analysis collected over a period of 12 months on a CEMS installed on the stack of a cement industry in Italy have been shown. The analysis was conducted using myleaf, the framework to gather and elaborate on information and actions. Due to the huge amount of data coming from the monitoring system (29 different parameters including stack and cabinet sensors) a multivariate analysis was first performed, in order to identify those parameters that have the greatest impact on the CEM system performance. Once identified most influential parameters, an univariate analysis has been performed. Although the plant and the instruments have shown no critical events during the analyzed period, this study allowed to obtain interesting and useful information that can be summarized as follows: - Those instrumental parameters that have to be monitor for preventive maintenance purposes have been identified: interferogram peak s intensity and position, X-Y-R laser signals, OBC board temperature, measuring cell s pressure, flue gas temperature. With regard to the interferogram peak, an algorithm able to predict when the signal may fall below the alarm threshold has been developed.

17 - Even those parameters that did not show a decay over time (e.g. laser current, IR source voltage and current) have demonstrated to be useful: indeed, they can provide information about the overall status of the CEMs by overlapping with other instrumental parameters. - Possibility of checking the effectiveness of maintenance after a system failure, by comparing the trend lines parameters before and after the maintenance intervention. - Possibility of follow, through the monitoring of the sample density allowed by the multivariate analysis, the trend of the parameters related to the system in its entirety: if such parameters tend over time to move out from the ellipse, it means that the system is going towards a probable failure. - Potentially myleaf could estimate the severity of failure so that, in case of minor troubles, the technician can lead the customer through remote assistance. For instance, such kind of assistance could be developed through a small camera placed on the helmet of the customer, so that the technician can see directly what happens into the plant in order to guide him during the maintenance. In the final analysis, results show that remote monitoring and data management allow to check in real time the parameters that affect the performance of the system and then to schedule maintenance activities efficiently while keeping the performance and the profitability of the system.

18 Bibliography [1] Jackson, J. E. (1991), A User's Guide to Principal Components, New York: John Wiley & Sons, Inc. [2] Willard, Merritt, Instrumental Methods of Analysis, Van Nostrand