Identification of intracellular calcium dynamics in stimulated cardiomyocytes

32nd Annual International Conference of the IEEE EMBS Buenos Aires, Argentina, August 31 - September 4, 2010 Identification of intracellular calcium dynamics in stimulated cardiomyocytes A. Vallmitjana, M. Barriga, Z. Nenadic, A. Llach, E. Alvarez-Lacalle, L. Hove-Madsen and R. Benitez Abstract We have developed an automatic method for the analysis and identification of dynamical regimes in intracellular calcium patterns from confocal calcium images. The method allows the identification of different dynamical patterns such as spatially concordant and discordant alternans, irregular behavior or phase-locking regimes such as period doubling or halving. The method can be applied to the analysis of different cardiac pathologies related to anomalies at the cellular level such as ventricular reentrant arrhythmias. I. INTRODUCTION There is an increasing number of studies that aim to establish relations between clinical conditions and physiological activity at the cellular level. This kind of research requires an interdisciplinary approach that combines knowledge and methods from different fields. Since most of the information at the cellular level is obtained by means of cell imaging techniques, novel image processing methods are needed in order to analyze, quantify and classify spatial and temporal patterns observed in life science areas such as neuroscience or cardiology [1]. In this context, calcium imaging is particularly relevant because calcium dynamics is a cell regulatory mechanism that plays an important role in many cellular processes such as muscle activation, gene expression or fertilization [2], [3]. In this work we present an automatic image processing method to analyze confocal calcium images of isolated cardiac myocytes. Cardiac myocytes are heart muscle cells that exhibit a variety of dynamical patterns due to the intracellular calcium dynamics [3]. The spatial and temporal distribution of intracellular calcium in cardiac myocytes determines the excitation-contraction coupling of the myocardium and is therefore a basic mechanism underlying heart function [4]. In particular, it is well known that high-frequency pacing of ventricular myocytes leads to the emergence of complex spatiotemporal patterns in the distribution of the intracellular calcium. The apparition of these complex dynamical regimes is a consequence of the nonlinear interplay between different cellular Ca 2+ control mechanisms [3], [5]. Irregular distribution of intracellular A. Vallmitjana and R. Benitez are with the Automatic Control Department, Universitat Politecnica de Catalunya (UPC), Barcelona, Spain. raul.benitez@upc.edu, alex.vallmitjana@upc.edu Z. Nenadic is with the Department of Biomedical Engineering, University of California, Irvine (USA). znenadic@uci.edu E. Alvarez-Lacalle is with the Applied Physics Department (UPC). enrical@fa.upc.edu L. Hove-Madsen, A. Llach and M. Barriga are with the Cardiovascular Research Center CSIC-ICCC and Cardiology Department, Hospital de Sant Pau (Barcelona, Spain). lhove@csic-iccc.org 978-1-4244-4124-2/10/$25.00 2010 IEEE 68 calcium may cause anomalies in the heart function such as T-wave alternans, ventricular fibrillation or conduction problems [6]. In particular, previous studies have established an interrelation between ventricular fibrillation and an overload in the intracellular calcium [7]. Similarly, the presence of spatially discordant alternans, characterized by an out-of-phase activity in different regions of the cell, is known to be related to the apparition of lethal arrhythmias [8] [10]. The purpose of this work is to present an analysis method that processes a sequence of fluorescence images of stimulated isolated myocytes and automatically identifies the spatiotemporal dynamics exhibited by the cell. The objective is to distinguish physiologically relevant regimes such as spatially concordant and discordant alternans, phase-locking oscillations or irregular patterns. The method uses a feature extraction technique that permits an effective characterization of the experimental sequence allowing for a robust identification of each regime. More specifically, an approach based in the Principal Component Analysis (PCA) is presented to detect the presence of spatial alternans in the experiment. Similar study that addresses this problem in the context of cardiac tissue patterns can be found in the recent literature [11]. The paper is organized as follows: In Section II we introduce the experimental data and provide a detailed description of the processing method. The main capabilities of the technique are described in Section III, where we evaluate its performance and report on several examples of the correct identification of different regimes. Finally, the potentialities of the method and an exposition of further improvements are discussed in Section IV. A. Data acquisition II. MATERIALS AND METHODS A total of 22 atrial myocytes were loaded with 2.5 µm fluo-4 for 15 minutes followed by wash and de-esterification for 30 minutes. The myocytes were stimulated intracellularly with an EPC-10 patch-clamp system (HEKA, Germany) as described in [12]. The sequences of confocal images were acquired at a frame rate of 100 Hz with a resonance scanning Leica SP5 AOBS confocal microscope. Ionic currents were recorded simultaneously with a HEKA EPC-10 amplifier. Synchronization of confocal images and current recordings was achieved using a Leica DAQ box and HEKA patchmaster software. Patch-master was used to design electro-

physiological protocols and to generate triggers for confocal image acquisition and event marking in the stimulation protocols. Local and global changes in cytosolic Ca 2+ levels were detected by quantifying fluo-4 fluorescence in selected regions of interest. The cardiomyocytes were analyzed at different stimulation rates with frequencies ranging from 0.25 to 2 Hz. This resulted in a set of 101 experimental sequences, each consisting in a sequence of N images of 512 140 pixels with a physical pixel size of 0.28µm. All the processing and analysis steps have been implemented in MATLAB TM (The Mathworks, Natick MA). The original fluorescence images (24-bit truecolor) are converted to grayscale intensity images by using a weighted sum of the R, G, and B components with weights [0.2989, 0.5870, 0.1140]. We refer to an experimental sequence of grayscale images as {Xij k }, where k = 1...N indexes the frame in the sequence and i =1...N x,j =1...N y specify a particular pixel in the image. In order to avoid the presence of static heterogeneities in the spatial distribution of the fluorescence, each pixel is normalized by subtracting its time average activity in the experiment. B. Feature extraction Fig. 1 describes the basic steps of the method, which includes feature extraction and classification. Feature extraction consists of two parts: On the one hand, we determine the temporal properties of the oscillations in the average fluorescence and its correspondence to the stimulation times. On the other hand, we analyze the experimental sequence in order to determine if the images present out-of-phase spatial heterogeneities. These two steps constitute a basis for peak detection and spatial analysis methods detailed below. Fig. 1. Schematic description of the method. 1) Peak detection: We first compute the average fluorescence cell activity in each frame F k = i,j Xk ij /(N xn y ),k =1...N, and we identify sequential pairs of local extrema corresponding to the peaks and valleys of F k. We then compute the mean and standard deviation of the peaks amplitude m a,σ a and of the intervals between consecutive peaks m i,σ i (inter-peak intervals). 69 The distribution of amplitudes is considered homogeneous if the variability of the peaks σ a is four times smaller than the noise in the signal σ 1. When the distribution of amplitudes is not homogeneous, alternating and irregular regimes are distinguished by testing for the presence of sustained oscillations in the peak amplitude. Similarly, an irregular behavior is identified when the variability in the inter-peak intervals exceeds a certain heuristic threshold σ i /m i > 0.6. Finally, the auto-correlation function of F k is used to determine the n:m correspondence between the calcium peaks and stimulation pulses. The previous procedure results in a set of four features, namely amplitude homogeneity, presence of alternance, irregularity of inter-peaks intervals and the n:m stimulation response. 2) Identification of spatial alternans: When the peak detection procedure detects the presence of an alternance in amplitude, an additional method is used in order to distinguish between spatially concordant or discordant alternans. To this extent, Principal Components Analysis (PCA) was used to identify the basic spatial modes in the experiment and to identify the existence of regions with an out-of-phase activity [13], [14]. In order to process the data, each image in the sequence Xij k,i = 1...N x,j = 1...N y was subtracted from its temporal mean and arranged as a d-dimensional column vector z k =[zk 1,z2 k,...,zd k ]T where d = N x N y. The whole experimental sequence was then represented by the d N matrix A =[z 1, z 2,, z N ]. The principal components are obtained by diagonalizing the d d covariance matrix AA T. In our case, since the dimension of the data d is much larger than the number of observations N (typical values are d 7 10 4, while N 2 10 3 ), we reduce the computational cost by using the fact that the largest N eigenvalues of AA T are the eigenvalues {λ 1,λ 2,...,λ N } of the N N matrix A T A [15]. The eigenvectors of AA T representing the spatial modes w can be then obtained from w = Av, where v are the eigenvectors of A T A. The main spatial mode in the experiment is found by reconstructing from the eigenvector w 1 associated with the largest eigenvalue λ 1. PCA reconstruction is achieved by projecting w 1 to the data matrix A, which results in an image representing the main spatial variability of the experimental sequence. The histogram of the reconstructed image is then divided in two regions A and B defined by the pixels above and below the average pixel intensity outside the cell (i.e. without calcium activity). The ratio between the pixel count in each region ρ = n B /n A defines a quantity that allows to identify the existence of regions presenting an out-ofphase activity in the sequence. Indeed, in the absence of spatial alternance the first order PCA reconstruction is homogeneous and the number of pixels in region B is low due 1 Noise is robustly estimated by the median absolute deviation of s k ˆσ = 1.4826 median( s k median(s k ) ), where s k = F k Fk d is a residual constructed from a denoised version Fk d obtained by applying a wavelet schrinkage method to the signal F k (Symmlet order 8, soft heuristic SURE threshold).

to background fluctuations in fluorescence (n B n A, i.e. ρ 0). When the sequence includes a spatially discordant alternant, the PCA projection captures the spatial variability by setting the pixels of the discordant region to negative values, therefore increasing the relative size of region B and consequently the value of ρ. A heuristic threshold as low as ρ =0.1 is proven to be sufficient to detect small spatial discordances. III. RESULTS A. Identification of Ca 2+ dynamical regimes The information obtained from the peak detection and PCA analysis provide a set of features that allow us to classify the experimental sequences into one of the following cases: 1) Normal dynamics: Normal cell response is characterized by a 1:1 stimulation response showing homogeneity in the peak amplitude and a spatial distribution of calcium activity. An example of this behavior is represented in Fig. 4a. As it can be seen, the cell responds to a train of stimulation pulses applied every 4 seconds by generating a calcium transient. This regime is the typical response of a healthy cell and is normally observed at low pacing frequencies. 2) Spatially concordant alternans: An example of spatially concordant alternans is depicted in Fig. 2, which shows a 1:1 stimulation response presenting an alternance in peak amplitudes. This temporal alternance appears in the whole cell without spatial inhomogeneities. 60 50 40 F/Fo 30 20 F/Fo 10 5 0 0 5 10 15 20 25 region 30 A c) region B 4 2 a) b) A 0 0 5 10 15 20 25 30 time (s) Fig. 3. Analysis of spatial alternance with PCA. A reconstruction from the most relevant eigenvector allows to identify two different regions A and B with alternating activities. of a calcium signal with a frequency different from the frequency imposed by external pacing. Fig. 4b shows an example of period-halving of the calcium signal with respect to the stimulation pulses, whereas Fig. 4c depicts a case in which every other stimulation pulse is blocked and evokes no calcium transient. F/Fo 15 10 5 0 0 5 10 15 b) 20 25 30 35 15 10 5 0 5 10 15 20 25 30 c) 10 a) B 10 5 0 0 5 10 15 20 25 30 35 40 time (s) 0 5 10 15 20 25 time (s) Fig. 2. Example of spatially concordant alternans at stimulation frequency 0.25 Hz. The whole cell responds to the stimulation pulses with alternating amplitudes. In all figures, F 0 corresponds to the background fluorescence of the quiescent cell and vertical marks indicate stimulation times. 3) Spatially discordant alternans: Spatially discordant alternans present different regions with out-of-phase activity in response to different stimulation pulses. In Fig. 3a out-ofphase regions A and B are presented. The corresponding average calcium signal of each region is shown in Fig. 3c, exhibiting an alternating behavior in the activity of each zone. The use of the PCA method becomes necessary since this regime cannot be distinguished from a spatially concordant alternant from the average cell activity (see Fig. 3b). 4) Phase-locking regimes: Phase-locking is a dynamical regime in which there is a n:m phase synchronization between stimulation pulses and peaks in the signal. In such cases, a nonlinear interaction between stimulation and calcium regulation mechanisms results in the appearance 70 Fig. 4. Normal cell response and phase-locking at pacing frequency 0.25 Hz. a) Normal dynamics b) Example of phase-locking 2:1 (period halving): The cell responds with two Ca 2+ transients every stimulation pulse. Note the correspondence between stimulation marks and signal peaks. c) Example of phase-locking 1:2 (period doubling): The cell responds with one Ca 2+ transient every two stimulation pulses (blocking). 5) Irregular dynamics: Irregular dynamics occur when either inter-peak intervals present significant variability (i.e., non-periodic behavior) or when peak amplitudes are highly heterogeneous presenting no alternance. In such cases, we observe dynamical regimes as the ones shown in Fig. 5. B. Performance evaluation To quantify the performance of the method, we analyzed the 101 experimental sequences and compared the classification results to those obtained by an expert. True and false positive rates (TPR, FPR) were computed for each of the four classification groups (Normal, phase-locking, Alternans -both concordant and discordant- and Irregular) as FPR =

F/Fo 10 5 60 40 20 0 0 2 4 6 8 10 12 14 16 18 20 22 b) 0 0 2 4 6 8 10 12 c) 10 5 a) 0 0 2 4 6 8 10 12 14 16 18 20 time (s) Fig. 5. Examples of irregular Ca 2+ transients at stimulation frequency of 1.33 Hz. TABLE I PERFORMANCE OF THE IDENTIFICATION METHOD Index Normal Phase-Locking Alternans Irregular TPR 92% 100% 80% 88% FPR 6% 14% 0% 16% Sample size 51 6 10 34 ratio of false positives over number of negatives and TPR = ratio of true positives over number of positives. Within the alternans group, the technique correctly distinguished all the cases presenting spatially discordant activity. IV. CONCLUSIONS AND FUTURE WORK A. Conclusions We have developed an automatic method for the identification of spatiotemporal regimes in a sequence of calcium fluorescence images in stimulated cardiomyocytes. The method distinguishes between spatially concordant and discordant alternating patterns and is able to identify phase-locking dynamics such as period doubling or halving as well as the presence of an irregular behavior. The technique can be used to obtain quantitative information about the dynamical response of the stimulated myocyte. In particular, it might be useful to characterize the sequence of bifurcations that the system undergoes as the pacing frequency is increased. Although the proposed method has been successfully applied to real experimental sequences, it would be necessary to quantify its performance and robustness under different signal-to-noise conditions. One of the straightforward improvements of the method is to substitute the PCA technique used for the identification of spatial alternans by an approach based on the use of Independent Components Analysis (ICA) [16]. This method would allow the decomposition of an experimental sequence in a set of statistically independent source signals associated with the alternating spatial modes. This might improve the overall method since PCA only identifies uncorrelated modes which are not necessarily statistically independent. Moreover, further dynamical information about the sequence may be obtained by using temporal and spatial phase synchronization techniques [5], [17], [18]. V. ACKNOWLEDGMENTS The authors acknowledge financial support by MICINN (Spain) under project DPI2009-06999. REFERENCES [1] J. Rittscher, R. Machiraju, and S. Wong, Eds., Microscopic Image analysis for life science applications, ser. Bioinformatics and Biomedical imaging. Artech House, 2008. [2] M. J. Berridge, M. D. Bootman, and H. L. Roderick, Calcium signalling: dynamics, homeostasis and remodelling, Nat Rev Mol Cell Biol, vol. 4, no. 7, pp. 517 29, Jul 2003. [3] J. P. Keener and J. Sneyd, Mathematical physiology, 2nd ed., ser. Interdisciplinary applied mathematics. New York, NY: Springer, 2009, vol. 8. [4] D. Bers, Cardiac excitation-contraction coupling, Nature, vol. 415, no. 6868, pp. 198 205, 2002. [5] S. H. Strogatz, Nonlinear Dynamics And Chaos: With Applications To Physics, Biology, Chemistry, And Engineering, 1st ed. Westview Press, 2001. [6] A. Karma and F. G. Jr, Nonlinear dynamics of heart rhythm disorders, Physics Today, vol. 60, no. 3, pp. 51 57, 2007. [7] E. Chudin, J. Goldhaber, A. Garfinkel, J. Weiss, and B. Kogan, Intracellular ca(2+) dynamics and the stability of ventricular tachycardia, Biophys J, vol. 77, no. 6, pp. 2930 41, Dec 1999. [8] M. A. Watanabe, F. H. Fenton, S. J. Evans, H. M. Hastings, and A. Karma, Mechanisms for discordant alternans, J Cardiovasc Electrophysiol, vol. 12, no. 2, pp. 196 206, Feb 2001. [9] D. Sato, Y. Shiferaw, A. Garfinkel, J. N. Weiss, Z. Qu, and A. Karma, Spatially discordant alternans in cardiac tissue: role of calcium cycling, Circ Res, vol. 99, no. 5, pp. 520 7, Sep 2006. [10] J. G. Restrepo and A. Karma, Spatiotemporal intracellular calcium dynamics during cardiac alternans, Chaos, vol. 19, no. 3, p. 037115, Sep 2009. [11] Z. Jia, H. Bien, and E. Entcheva, Detecting space-time alternating biological signals close to the bifurcation point, IEEE Trans Biomed Eng, vol. 57, no. 2, pp. 316 24, Feb 2010. [12] L. Hove-Madsen, C. Prat-Vidal, A. Llach, F. Ciruela, V. Casadó, C. Lluis, A. Bayes-Genis, J. Cinca, and R. Franco, Adenosine a2a receptors are expressed in human atrial myocytes and modulate spontaneous sarcoplasmic reticulum calcium release, Cardiovasc Res, vol. 72, no. 2, pp. 292 302, Nov 2006. [13] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, 2nd ed. John Wiley and Sons, Inc., 2001. [14] B. Ghanem and N. Ahuja, Phase PCA for dynamic texture video compression, in IEEE International Conference on Image Processing, 2007. [15] G. Blanchet and M. Charbit, Digital signal and image processing using MATLAB. ISTE-Wiley, 2006. [16] J. V. Stone, Independent Component Analysis: A tutorial introduction. The MIT Press, 2004. [17] M. Palus, Detecting phase synchronization in noisy systems, Physics Letters A, vol. 235, no. 4, 1997. [18] M. G. Rosenblum, A. S. Pikovsky, and J. Kurths, Phase synchronization of chaotic oscillators, Phys. Rev. Lett., vol. 76, no. 11, pp. 1804 1807, 1996. 71

Conditioning Data for Condition Assessment of a Power Transformer Roberto Villafáfila-Robles 1, Marta Rodríguez 1, Pau Lloret 1, Andreas Sumper 1,2, Samuel Galceran-Arellano 1 1 Centre of Technological Innovation in Static Converters and Drives (CITCEA) Universitat Politècnica de Catalunya (UPC) E.U. d Enginyeria Tècnica Industrial de Barcelona, Electrical Engineering Department Comte d Urgell, 189. 08036 Barcelona (Spain) e-mail: roberto.villafafila@citcea.upc.edu 2 Catalonia Institute for Energy Research (IREC) Barcelona (Spain) Abstract Utilities have to guarantee a proper condition of network components in order to meet with regulatory and society demands regarding reliability and quality of power supply while optimizing costs. Maintenance strategies have evolved to cope with this issue. Condition Based Maintenance (CBM) strategy permits to adapt the maintenance actions to condition of the equipment. It is mainly used for critical equipment like power transformers. If an on-line monitoring system is used, the actual condition of the assets can be estimated. Such system consists of a set of sensors for acquiring condition related parameters and techniques/tools that process and analyze the data in order to assess its condition. However, anomalous data may appear due to a malfunction of monitoring system and may lead to errors when in data interpretation. Then, in order to overcome this issue, conditioning of such data is needed previously to analyse them. When the monitored data is refined, the condition can be estimated through models. A conditioning data process is presented for a case study of a power transformer in service. Furthermore, data mining process for obtaining behaviour patterns is also introduced. Keywords: Condition monitoring system, Condition assessment, Conditioning Data, Condition Based Maintenance I. INTRODUCTION Asset management has become one of the main activities for utilities due to liberalization of electric sector. This environment needs for new strategies in operation and maintenance activities in order to reduce their costs while improving reliability and quality of power supply in order to meet with regulatory frames and society demands. Furthermore, the risk is likely to increase when optimizing technical and economical resources if financial interests are above the actual condition of the assets and not at the same level. The condition of assets is guaranteed through maintenance actions. Such actions can be grouped in different strategies depending of the criticality of the asset, its cost and available spare parts. There are four main maintenance strategies with the following characteristics [1]: Corrective maintenance (CM): there is no inspection or maintenance until breakdown. Time Based maintenance (TBM): there is a fixed time intervals for inspections and maintenance. Condition Based maintenance (CBM): there is continuous or occasional monitoring and the maintenance is performed when required. Reliability Centred maintenance (RCM): there is a priority list obtained from a connection of condition and failure effects that permits risk management. Utilities have been mainly performing maintenance of their assets as a combination of CM and TBM strategies, depending on the network component. However, a CM plan will have a significant impact on power system operation if critical component failures. On the other hand, TBM plans might over-maintain young equipment whereas infra-maintain ones close to their end-of-life. Thus, there is a shift towards a CBM approach for critical equipment, like power transformers, in order to avoid damages of network components by means of detecting faults at incipient stage. As it is no possible to measure directly the time to failure of any network component, 1

such time is predicted by means of monitoring parameters that can provide an approximation of actual condition and ageing process after the corresponding analysis. A step forward in maintenance strategies after CBM is RCM. This last plan considers, apart from actual condition, other facts like resource constraints and power quality indices to prioritize the maintenance orders. However, in order to set up last two maintenance strategies, utilities require a high financial effort for deploying the related systems, and qualified and experienced staff able to manage and take advantage of such systems. The monitoring of condition related parameters of equipment can be done through both on-line and off-line methods. In order to carry out an on-line monitoring system, it is necessary to install sensors that continuously acquire the data from the monitored network component and information and communication system that transmits and storages such data. Then, these data are accessible for a later analysis. However, the installation of sensors represents an important drawback for equipment in service. Off-line monitoring methods can overcome this problem by checking equipment that should be out of service. However, such measurements might be done too late for preventing damages or might not provide useful results. Any action within a CBM strategy, like alarms, maintenance or replace orders, depends on assess of the condition of the equipment and later diagnosis. On one hand, some monitoring techniques use monitored data in standard models, like thermal models defined at IEEE Std. C.57.91 and IEC-354. However, such models consider parameters that have to be calculated for each one. On the other hand, other monitoring methods require power transformer s fingerprint that is used as reference in later analysis to determine the evolution, like Frequency Response Analysis (FRA). However, these techniques need qualified staff to perform the test and assess the results. Thus, determining the condition of a power transformer and the limits to raise the alarm is a cumbersome task. Power transformers are a key component in power systems and utilities are doing huge efforts for avoiding damages in such equipment by deploying CBM plans for them. The techniques used for condition monitoring and condition assessment for power transformers can be found in [2]. As it has been already mentioned, on-line monitoring for a CBM implies two steps: a data acquisition system that gets the value of condition related parameters of the equipment and techniques/tools that process and analyze such data in order to assess its condition. The main outcome is to detect incipient faults and perform proper actions to reduce the damage and recover a good state-of-health of the equipment However, conditioning the data from monitoring system is needed in order to get useful information and remove erroneous data. If this process is not performed properly, it could conduct to wrong results. A mistake in the assessment of condition can lead to loss of both the equipment and significant amount of money. This paper deals with conditioning monitored data for estimating the condition of a transformer based on a case study. II. PILOT PLANT A condition monitoring pilot plant has been deployed according to methodology described in [3]. The pilot plant is shown in Figure 1. It consists in 66/25 kv 30 MVA power transformer and substation circuit-breakers. The description of the condition monitoring system: sensors, data acquisition and warehouse systems, and communication systems are described deeper in [4] and [5]. The parameters of the active part of the power transformer that are monitored and the sensors are listed in Table I. The values of such parameters are acquired continuously and a pre-process is done before they are stored in the data base. The storage of these parameters is synchronous: instant values are aggregated in the average every 15 minutes and such average and the maximum and the minimum for each quarter of an hour are recorded. Date and time are recorded with each measure. Table I. Power transformer monitored variable Monitoring parameter Sensor Upper oil temperature Pt100 Gases dissolved in oil Hydran M2 Oil humidity Vaisala MMT318 Lower oil temperature Pt100 High-voltage 3-phase currents Current transformer High-voltage 3-phase voltages Voltage transformer 2

III. CONDITIONING MONITORED DATA Figure 1. Monitoring pilot plant The data base stores data that can be used for assessing the condition of power transformer. A first step is to plot such data. Figure 2 shows monitored data of transformer temperature from the monitoring system for the same month in two different years. It can be seen that some data is missed or present a value equal to zero. Therefore, a conditioning procedure is needed to identify the cause of this situation and extract accurate information that permits estimate the condition of the power transformer in order to specify the proper maintenance actions if needed. The proposed methodology is shown in Figure 3 and is described next. It has two parts: finding wrong data and generate patter of behaviour. The objective is to obtain the set of data free of anomalous values and create for each monitored parameter a behaviour pattern to identify changes or trends that conduct to an unwanted situation Evolution of mean temperature values Evolution of mean temperature values ºC 60 50 40 30 20 10 0 01/04/2008 02/04/2008 03/04/2008 04/04/2008 05/04/2008 06/04/2008 07/04/2008 08/04/2008 10/04/2008 11/04/2008 12/04/2008 13/04/2008 14/04/2008 15/04/2008 16/04/2008 17/04/2008 19/04/2008 20/04/2008 21/04/2008 22/04/2008 23/04/2008 24/04/2008 Time Top-layer Bottom-layer Ambient 60 50 40 30 20 10 0 01/04/2009 02/04/2009 04/04/2009 06/04/2009 08/04/2009 09/04/2009 10/04/2009 11/04/2009 12/04/2009 13/04/2009 14/04/2009 15/04/2009 17/04/2009 18/04/2009 19/04/2009 20/04/2009 21/04/2009 22/04/2009 23/04/2009 24/04/2009 26/04/2009 28/04/2009 29/04/2009 30/04/2009 Figure 2. Temperature measurements in April 2008 (left) and 2009 (right) ºC Time Top-layer Bottom-layer Ambient a. IDENTIFYING WRONG DATA On-line monitoring systems might have some malfunction that cause that anomalous data is inserted in the data base. Such abnormal data have not to be taken into consideration for assessing the condition. The origin of these inaccuracies is misoperation of some of its components like sensors and disfunction of communication and software systems, as for example a damaged sensor, loss of communication between sensors and data base due to the cable is broken, and writing failure when inserting in data base. In order to cope with these sources of errors, stored values of monitored parameters are asked next questions: 3

o Is the number of data expected? This doubt discovers missing intervals as the storage of monitored data is carried out in constant time intervals. o Is there a date without measure? This enquiry notices that a measurement is not recorded in the data base. o Are there data with zero value? This issue detects an error in stored data, although a null value in current and voltage could mean that the transformer is out of service. As a result of each question, a list with the detected wrong data is created and stored. The exact cause of misoperation of the on-line monitoring system can be determined by analysing the data lists generated after the question. After the whole set of data goes through the questions, the appropriate data is available for obtaining behaviour pattern of each parameter. Figure 3. Algorithm for conditioning monitored data 4

b. GENERATION OF BEHAVIOUR PATTERNS The behaviour patterns are found through data mining process applied to free error set of monitored data. Data mining has been selected due to it is an efficient technique to obtain useful information from cleaned large amount of data. There are different techniques for performing data mining: neuronal networks, decision trees, genetic algorithms, clustering, linear regression, statistics, etc. Statistical analysis has been selected to derive the behaviour patterns for watching the evolution of the condition of the power transformer. This technique consists of adjusting the data to a statistical distribution model. A data distribution fit-test determines the suitable model. Before performing such tests, the influence of the season and time of day have to be considered. Then, refined data are separated in winter (from December to February), spring (from March to May), summer (from June to August) and autumn (from September to November); and for each season, the data is considered hourly. The fit-tests have been carried out according to previous conditions and the normal distribution fits with the refined data, as Figure 4 shows for top-layer temperature. Figure 4. Distribution fit-tests (with Minitab ) for top-layer temperature: exponential (upper left), Weibull (lower left), normal (upper right) and log-normal (lower right) Therefore, each parameter has four behaviour patterns that each one consists of a daily model made of 24 normal distributions, one for each hour of the day. Figure 5 depict the behaviour pattern of top-layer temperature for spring, where hourly means are connected by a continuous line, and the upper and lower lines limit the confidence interval of 95.44% ( ±2 ). For deriving this pattern, the wrong data that Figure 2 shows in April 2009 has not been taking into account and do not affect it. The models are stored in the data base and when the monitoring system acquires new raw data, such data is firstly refined and later is used for updating the corresponding pattern. The behaviour patterns permit to assess the evolution of monitored parameters and evaluate the condition of the power transformer by means of comparison and correlation between the parameters. Figure 5. Spring top-temperature behaviour pattern. Continuous line: mean. Dot-point line: upper limit. Dot line: lower limit 5

IV. CONCLUSIONS Power transformers are an important asset in power systems. Monitoring of power transformers permits to estimate their condition and life expectancy. Although degradation process of insulation materials and failure modes are known, the assessment of their ageing and time to failure is hard difficult. On-line monitoring systems help to estimate current condition of power transformers. However, raw data might present anomalous values due to malfunction of the monitoring system and if these are not identified, incorrect conclusions could appear in later analysis. In order to overcome such situation, a refining stage previous to condition analysis is needed. A conditioning monitored data process for an on-line monitoring pilot plant system has been described. This process permits to derive behaviour patterns to identify changes or trends that might conduct to an unwanted condition of power transformer. The patterns have been derived from cleaned monitored data using statistics data mining techniques, namely normal distribution considering season and each hour of the day. REFERENCES [1] Joachim Schneider, Armin J. Gaul, Claus Neumann, Jurgen Hografer, Wolfram Well ow, Michael Schwan, Armin Schnettler, Asset management techniques, International Journal of Electrical Power & Energy Systems, Volume 28, Issue 9, Selection of Papers from 15th Power Systems Computation Conference, 2005 - PSCC'05, November 2006, Pages 643-654, ISSN 0142-0615, DOI: 10.1016/j.ijepes.2006.03.007. [2] Ahmed E.B. Abu-Elanien, M.M.A. Salama, Asset management techniques for transformers, Electric Power Systems Research, Volume 80, Issue 4, April 2010, Pages 456-464, ISSN 0378-7796, DOI: 10.1016/j.epsr.2009.10.008. [3] Velasquez, J.L.; Villafafila, R.; Lloret, P.; Molas, L.; Galceran, S.; "Guidelines for the implementation of condition monitoring systems in power transformers," Advanced Research Workshop on Transformers 2007, ARWtr2007, vol., no., pp.1-6, 29-31 Oct. 2007. [4] Velasquez, J.L.; Villafafila, R.; Lloret, P.; Molas, L.; Sumper, A.; Galceran, S.; Sudria, A.; "Development and implementation of a condition monitoring system in a substation," International Conference on Electrical Power Quality and Utilisation, 2007, EPQU 2007, vol., no., pp.1-5, 9-11 Oct. 2007. [5] Lloret, P.; Velasquez, J.L.; Molas-Balada, L.; Villafafila, R.; Sumper, A.; Galceran-Arellano, S.; "IEC 61850 as a flexible tool for electrical systems monitoring," 9th International Conference on Electrical Power Quality and Utilisation, 2007, EPQU 2007, vol., no., pp.1-6, 9-11 Oct. 2007. ACKNOWLEDGEMENT The pilot plant project has awarded with Endesa s R+D+i international prize NOVARE 2005 on distribution networks in the category of Power Quality and Reliability by the project: 'Substation monitoring for predictive maintenance'. 6

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Formación y evaluación de la competencia en habilidad espacial Jordi Torner, Francesc Alpiste Penalba, Miguel Brigos Hermida Urgell 187, Barcelona, 934137398, 934017800 jordi.torner-ribe@upc.edu Resumen Diversos estudios señalan la habilidad espacial como una variable clave en los estudios de Ingeniería Industrial. Es fundamental para la actividad proyectual del ingeniero ya que resulta vital en el diseño de proyectos. Entre otros factores, se correlaciona con buenos resultados académicos y con facilidad de aprendizaje de sistemas de información y herramientas informáticas. Asimismo, el nuevo escenario creado por el EEES nos conduce a la definición y medición competencias, entre las cuales la integraremos. En el presente artículo planteamos la estrecha relación que une el desarrollo de esta habilidad con el trabajo con software de modelado de sólidos en 3D. El estudio se realiza con 812 estudiantes de 1er año de Ingeniería Industrial de la Universidad Politécnica de Catalunya, analizando la evolución de las puntuaciones obtenidas a través de los test DAT-SR y MRT, antes y después de la asignatura de diseño asistido por ordenador. Palabras Clave: competencias; EEES; habilidad espacial. Abstract Many studies show that spatial ability is a key factor in engineering studies. It is essential for the engineer in sketching activity and vital on projects design. Among other factors, it is correlated with brilliant academic results and capability on learning information systems and software. Besides, the new scenario created by EEES drives us to the competences definition and evaluation. In this paper we show the big relationship between this ability development with 3D solid modelling software. This study is made with 812 first year Engineering students at UPC-Barcelona Tech, analyzing the scores evolution over DAT-SR and MRT tests, before and after computer aided design subject. Keywords: competences; EEES; spatial ability. 1. Introducción La inteligencia humana se pone de manifiesto en el nivel de desarrollo de ciertas habilidades (verbal, numérica, espacial, etc.). Diversos autores destacan la importancia de la habilidad espacial (HE) en los procesos de diseño en Ingeniería y proponen estrategias didácticas para favorecer su desarrollo entre los estudiantes. El desarrollo de la habilidad espacial forma parte del currículum de la Ingeniería Gráfica desde hace largo tiempo [1]. En los últimos años, el interés ha XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria ido creciendo debido a las novedades y el impulso tomado desde la informática gráfica. Su valor reside básicamente en la relación entre la HE con el diseño y con la comunicación gráfica. El concepto de HE cubre un amplio abanico de funciones cognitivas. En la actualidad existen multitud de tests y pruebas que permiten abordar los diferentes componentes de dicha habilidad. Este hecho provoca que el concepto quede fragmentado en múltiples sub-factores y resulta complicado encontrar una definición aceptada de forma unánime por toda la comunidad científica. No obstante, encontramos 2 componentes básicos de la habilidad del que derivan los demás, aceptados por la comunidad científica [1]: Visión espacial: habilidad de manipular un objeto en un espacio 3D imaginario creando representaciones del objeto desde diferentes puntos de vista. Orientación espacial: se refiere a la capacidad para controlar el espacio de nuestro entorno y predecir el movimiento y la posición de los objetos. Un ingeniero debe ser capaz de resolver gráficamente la representación de estructuras y sistemas complejos en el desarrollo de su trabajo. Por lo que necesariamente la HE es útil y puede llegar a ser clave en el desarrollo de proyectos de ingeniería, tal y como apuntan diversos estudios [2,3]. En las primeras fases del diseño de proyectos es fundamental solventar con rapidez problemas en los que el razonamiento espacial juega un papel decisivo, por ejemplo, en la fase de croquización. Por otra parte, la HE se ha reconocido como factor determinante en la predicción de éxito en diversas áreas, especialmente en las áreas tecnológicas [4]. Es decir, se han establecido correlaciones positivas entre la HE y los resultados académicos de los estudiantes en ingeniería. Se han establecido correlaciones positivas con la capacidad de aprendizaje de aplicaciones informáticas, herramientas de CAD, en el diseño de Bases de Datos o en el desarrollo de estructuras moleculares [5]. El aprendizaje de una herramienta profesional de CAD en los estudios de Ingeniería Industrial se hace cada vez más necesaria debido, entre otros factores, a la demanda XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria del mercado laboral. En consecuencia la gran mayoría de universidades y escuelas técnicas utilizan una herramienta de CAD en los primeros cursos de las ingenierías. Varios autores [6,7] han demostrado que el uso de herramientas CAD puede potenciar el desarrollo de la visión espacial. En resumen, la HE se configura como: Competencia básica en el currículum del ingeniero. Fundamental para la actividad proyectual: Resulta vital en el diseño y desarrollo de proyectos. Se correlaciona con buenos resultados académicos y con facilidad de aprendizaje de sistemas de información y herramientas informáticas Necesarias para resolver gráficamente la representación de estructuras y sistemas complejos en el desarrollo de su trabajo. Factor determinante en la predicción de éxito en diversas áreas, especialmente en las áreas de ciencias y tecnológicas (correlaciones positivas con resultados académicos de los estudiantes en ingeniería). Se han establecido correlaciones positivas con la capacidad de aprendizaje de aplicaciones informáticas, herramientas de CAD, en el diseño de BBDD o en el desarrollo de estructuras moleculares. Relación entre la HE i la habilidad para trabajar con sistemas de información informáticos (navegación por menús jerárquicos y bases de datos, portales de e-learning, sistemas de almacenamiento de información y en general todo tipo de espacios web). Debido a todos estos condicionantes, este trabajo pretende desarrollar un modelo que permita evaluar la HE de los estudiantes de ingeniería de la asignatura de primer curso Expresión Gráfica y diseño asistido por ordenador. XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria El modelo contará con los procedimientos y con los indicadores necesarios, permitirá ponderar las principales variables y orientar sobre las acciones a introducir de mejora de la práctica docente. Nuestro objetivo es comprobar si el uso de una herramienta de modelado de sólidos 3D, como Solidworks, desarrolla la HE en los estudiantes. Para ello se realiza la pasación de 2 tests de HE al inicio y final del cuatrimestre y se comprobará si existen diferencias significativas entre las puntuaciones obtenidas antes y después de las clases. De esta manera podremos estudiar si la intervención realizada en las clases de la asignatura produce un entrenamiento de la HE. El estudio se realiza en un momento de cambios importantes ya que se ha procedido a la adaptación de la asignatura Expresión Gráfica y Diseño asistido por ordenador al modelo acordado en Bolonia. La integración de las universidades en el Espacio Europeo de Educación Superior (EEES) nos conduce a modificar la estructura, los contenidos y el modelo de enseñanza-aprendizaje de nuestros programas de formación. Este hecho nos abre un nuevo eje en la investigación. Este escenario nos lleva a la definición de competencias: Las actividades formativas se orientan a la adquisición de competencias específicas de cada asignatura, adoptándose un enfoque formativo-práctico. Y a la evaluación de resultados: Se requiere la evaluación de los resultados obtenidos en el proceso en términos de competencia, intentando acercar el perfil profesional al académico observando los conocimientos y habilidades necesarios en el mundo laboral. La integración al espacio Europeo nos conduce a definir las competencias específicas que definirán la asignatura de Expresión Gráfica y DAO. Tal y como se ha comentado una de las competencias más importantes en la figura del Ingeniero y de la asignatura es la HE. Por lo tanto definir y evaluar dicha competencia se convierte en otro objetivo importante de nuestra investigación. XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria 2. Objetivos y metodología El objetivo último del estudio es desarrollar un modelo que permita evaluar la HE en los estudiantes de Ingeniería Industrial y que permita a su vez evaluar las estrategias y los métodos de la programación y su relación con la HE. Figura 1. Esquema de investigación En el desarrollo de este trabajo se ha analizado la asignatura de primer curso Expresión Gráfica y Diseño asistido por ordenador de la UPC. Mediante la utilización del modelo propuesto, se quiere comprobar si la metodología didáctica y las actividades realizadas en la asignatura colaboran en un desarrollo significativo de la HE en los estudiantes. En referencia a la detección de HE mediante test se estudiaran las principales soluciones utilizadas. De entre ellas, se elegirá la tipología de test que se adapte mejor al objetivo de nuestro estudio. XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Con este objetivo se pasarán 2 test (DAT-SR y MRT) de HE al inicio y final del cuatrimestre y se comprobará si existen diferencias significativas entre las puntuaciones obtenidas antes y después de las clases mediante técnicas estadísticas. Figura 2. Rotación de figuras (basado en MRT) 3. Competencia en HE Con el EEES la definición y evaluación de competencias adquiere un papel relevante. La competencia es la habilidad aprendida para llevar a cabo una tarea, deber o rol adecuadamente. Un alto nivel de competencia es un prerrequisito de buena ejecución. Navío [8] apunta que las competencias profesionales son un conjunto de elementos combinados que se integran atendiendo a una serie de atributos personales tomando como referencia las experiencias personales y profesionales y que se manifiestan mediante determinados comportamientos o conductas en el contexto de trabajo. Destacan entre otros el trabajo de Moon [9] para la programación de la asignatura y el de Urraza [10] que nos propone un modelo de competencias de la asignatura en el que la HE queda integrada: XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Tabla 1.. Competencias específicas de Expresión Gráfica y DAO y relación con las competencias trasversales implicadas. COMPETENCIAS ESPECÍFICAS C. TRANSVERSALES T.I: Instrumentales; T.P: Interpersonales T.S: Sistémicas COMPETENCIAS RELACIONADAS CON LOS CONCEPTOS Y CONOCIMIENTOS BÁSICOS C.1 Comprender, gestionar y aplicar un soporte de conocimientos sobre los fundamentos y normalización del Dibujo de Ingeniería Industrial, plataforma necesaria para abordar los problemas de ingeniería gráfica. T.I.2. Capacidad de análisis y síntesis T.I.3. Capacidad de gestión de la información T.I.5. Conocimientos básicos de la profesión T.S.2. Aprendizaje autónomo C.2 Aplicar con destreza los programas de DAO, que hacen que el ordenador se constituya en una herramienta didáctica, precisa y rápida, para la confección de la base documental de los objetos que deben de ser representados desde la perspectiva de los conocimientos del Dibujo de Ingeniería. T.I.6. Conocimientos de informática T.S.2. Aprendizaje autónomo COMPETENCIAS RELACIONADAS CON EL APRENDIZAJE CONSTRUCTIVISTA C.3 Gestionar y aplicar la capacidad espacial utilizando como soporte la croquización, dentro de un marco de desarrollo estrategias cognitivas que ayuden a la visualización tridimensional de los objetos técnicos. T.I.1. Resolución de problemas T.S.2. Aprendizaje autónomo C.4 Interpretar y realizar planos normalizados del Dibujo de Ingeniería Industrial. T.I.1. Resolución de problemas T.S.1. Capacidad de aplicar los XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria conocimientos a la práctica T.S.2. Aprendizaje autónomo C.5 Aplicar el conocimiento procedimental en la resolución de los problemas de la Geometría Constructiva orientados a la representación de superficies. T.I.1. Resolución de problemas T.S.1. Capacidad de aplicar los conocimientos a la práctica T.S.2. Aprendizaje autónomo C.6 Aplicar las habilidades de investigación y la creatividad en la introducción al diseño industrial. T.I.1. Resolución de problemas T.S.1. Capacidad de aplicar los conocimientos a la práctica T.S.2. Aprendizaje autónomo T.S.3. Creatividad T.S.5. Habilidades de investigación C.7 Gestionar las fuentes de información, exponiendo y justificando de forma gráfica, oral y escrita los aspectos relacionados con las ideas de diseño y con la interpretación y realización de los documentos de Ingeniería. T.I.4. Capacidad de organización y planificación T.I.7. Comunicación gráfica, oral y escrita C.8 Trabajo en equipo que facilite el desarrollo de los conocimientos con un intercambio cultural crítico y responsable. T.P.1. Trabajo en equipo T.P.2. Capacidad de crítica y autocrítica XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria 4. Modelo Definimos un Modelo para el desarrollo de la HE en la Expresión Gráfica. El objetivo del modelo es disponer de recursos de mejora docente a partir del estudio de la HE. El modelo permite el control de variables que afectan la HE y facilita su medida Pre y Post curso. Además, el modelo establece relaciones entre las metodologías didácticas, los resultados académicos y la satisfacción de los estudiantes. Figura 3. Modelo para el desarrollo de la HE en la Expresión Gráfica La actividad se centra en la programación de la asignatura de Expresión Gráfica y Diseño Asistido por Ordenador (EGDAO) y en el estudio de las habilidades espaciales que se desarrollan en ella. XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010

ETS de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Se aporta un modelo para medir la mejora de la habilidad espacial, qué es una competencia básica de los ingenieros. Para ello, se describen las variables que afectan la HE y se propone un sistema de depuración de las mismas que, además, orienta en las acciones didácticas a tomar para mejorar la HE Los estudios estadísticos realizados permiten obtener valores cuantitativos que pueden ser utilizados como referencia para los indicadores de calidad. Además contribuyen en la determinación de la fiabilidad de las encuestas realizadas. Se determina la correlación entre los valores de HE y los resultados académicos obtenidos a partir de las evaluaciones de las principales actividades didácticas realizadas. Esta correlación nos permite determinar la influencia de las metodologías docentes utilizadas en la mejora de HE y nos orienta acerca de la selección de actividades más eficaces. Finalmente, todos los datos analizados revierten en la toma de decisiones para incidir en la mejora de la calidad docente, toda vez que disponemos de un conjunto de métodos y herramientas con las que obtener y comparar los registros con los indicadores de referencia utilizados en un proceso de mejora continua. 5. Conclusiones y líneas futuras de investigación De todas las variables analizadas en el estudio, se identifican mediante el análisis de los resultados, las siguientes variables determinantes en las puntuaciones de HE: Uso de software de CAD: se aprecian diferencias significativas en los alumnos con experiencia en este tipo de programas. Especialidad: encontramos diferencias importantes entre especialidades, especialmente en química, que obtiene las medias más bajas. La relación más fuerte se encuentra entre del DAT inicial y la prueba DAO2 dedicada a la geometría del espacio. Por lo tanto, se propone potenciar las actividades relacionadas con la geometría del espacio para maximizar el desarrollo de la HE. XVIII Congreso Universitario de Innovación Educativa en las Enseñanzas Técnicas Escuela Técnica Superior de Ingenieros Industriales y de Telecomunicación. Universidad de Cantabria Santander, 6 a 9 de julio de 2010