Optimal Combination Of Sensors Research Articles

Improvement in Identification Accuracy of a Failure Diagnostic System for a Reusable Rocket Engine

As a technology for safe and efficient operation of reusable rockets, we are developing failure diagnosis technology for reusable rocket engines. In order to follow the changes in rocket engine operating conditions, a failure diagnostic method which monitors an error vector: the difference between the predicted and measured values of the sensors was developed. The method contains anomaly detection by Mahalanobis distance and failure identification by support vector machines (SVMs). In this report, combinations of monitoring sensors of SVMs for each failure mode were optimized by using design of experiments. By using optimal sensor combinations, the F-score of SVMs were improved in all failure modes. From the results of the orthogonal table experiments, it was supposed that sensors which show the difference in failure modes are important to distinguish failure modes. In addition, a failure classifier combined with the optimized SVMs for each failure mode was developed and demonstrated. The performance of the combined failure classifier with the optimal sensor combination was mostly greater than that with all sensors. However, degradation of the classification performance was also obtained. It is necessary to consider how integrate the results of SVMs which are optimized individually.

A data-driven adaptive algorithm and decision support design of multisensory information fusion for prognostics and health management applications

Multisensory systems play a critical role in prognostics and health management (PHM), and utilise the information from multi-device synchronous measurements for fault diagnosis and predictive maintenance. But it is not suitable for specific systems with limited bandwidth and energy reservoirs since the increased sophistication of measurement devices requires more computation and power resources. This research explores a data-driven analytical framework for multisensory system analysis and design in PHM. The proposed framework provides the optimal subset of reliable sensors to make trade-offs between accuracy demands and system constraints. The integration definition for function modelling method is adopted for modelling and functional analysis of the proposed framework. An adaptive signal conversion algorithm is designed to process the data from all reliable sensors in the system. The convolutional neural network with residual learning is built for automatic feature extraction. Combined with the evaluation rules and expert knowledge, performance analyses are obtained, including qualitative results, fault diagnosis, and the optimal sensor combination. An open-source bearing dataset of the multisensory system with five measurements is conducted to demonstrate the effectiveness and feasibility of the proposed framework.

Metrological aspects of determining the electron concentration and the conductivity of a melt using a conductivity sensor

For radiophysics, research into plasma physics is of great interest. During the movement of a hypersonic aircraft, the main physical phenomenon, which radically affects the conditions for the propagation of electromagnetic waves in its vicinity is the formation of a plasma shell around the apparatus due to the ionization of neutral particles. The resulting plasma around a hypersonic aircraft has a very high concentration of electrons and prevents the propagation of electromagnetic waves in space, partially absorbing them and reflecting them from its surface. When measuring the electron concentration, it is necessary to take into account the measurement error. In this article, the error is measured on the basis of expressions obtained for a transmission line whose longitudinal dimension exceeds the wavelength propagating in it. Mathematical model for finding the error in determining the electron concentration and melt conductivity is considered, and an analysis of the optimal combination of sensors is carried out.

Estimating dry biomass and plant nitrogen concentration in pre-Alpine grasslands with low-cost UAS-borne multispectral data – a comparison of sensors, algorithms, and predictor sets

Abstract. Grasslands are an important part of pre-Alpine and Alpine landscapes. Despite the economic value and the significant role of grasslands in carbon and nitrogen (N) cycling, spatially explicit information on grassland biomass and quality is rarely available. Remotely sensed data from unmanned aircraft systems (UASs) and satellites might be an option to overcome this gap. Our study aims to investigate the potential of low-cost UAS-based multispectral sensors for estimating above-ground biomass (dry matter, DM) and plant N concentration. In our analysis, we compared two different sensors (Parrot Sequoia, SEQ; MicaSense RedEdge-M, REM), three statistical models (linear model; random forests, RFs; gradient-boosting machines, GBMs), and six predictor sets (i.e. different combinations of raw reflectance, vegetation indices, and canopy height). Canopy height information can be derived from UAS sensors but was not available in our study. Therefore, we tested the added value of this structural information with in situ measured bulk canopy height data. A combined field sampling and flight campaign was conducted in April 2018 at different grassland sites in southern Germany to obtain in situ and the corresponding spectral data. The hyper-parameters of the two machine learning (ML) approaches (RF, GBM) were optimized, and all model setups were run with a 6-fold cross-validation. Linear models were characterized by very low statistical performance measures, thus were not suitable to estimate DM and plant N concentration using UAS data. The non-linear ML algorithms showed an acceptable regression performance for all sensor–predictor set combinations with average (avg; cross-validated, cv) Rcv2 of 0.48, RMSEcv,avg of 53.0 g m2, and rRMSEcv,avg (relative) of 15.9 % for DM and with Rcv,avg2 of 0.40, RMSEcv,avg of 0.48 wt %, and rRMSEcv, avg of 15.2 % for plant N concentration estimation. The optimal combination of sensors, ML algorithms, and predictor sets notably improved the model performance. The best model performance for the estimation of DM (Rcv2=0.67, RMSEcv=41.9 g m2, rRMSEcv=12.6 %) was achieved with an RF model that utilizes all possible predictors and REM sensor data. The best model for plant N concentration was a combination of an RF model with all predictors and SEQ sensor data (Rcv2=0.47, RMSEcv=0.45 wt %, rRMSEcv=14.2 %). DM models with the spectral input of REM performed significantly better than those with SEQ data, while for N concentration models, it was the other way round. The choice of predictors was most influential on model performance, while the effect of the chosen ML algorithm was generally lower. The addition of canopy height to the spectral data in the predictor set significantly improved the DM models. In our study, calibrating the ML algorithm improved the model performance substantially, which shows the importance of this step.

Modeling and Fault Detection of Brushless Direct Current Motor by Deep Learning Sensor Data Fusion.

Only with new sensor concepts in a network, which go far beyond what the current state-of-the-art can offer, can current and future requirements for flexibility, safety, and security be met. The combination of data from many sensors allows a richer representation of the observed phenomenon, e.g., system degradation, which can facilitate analysis and decision-making processes. This work addresses the topic of predictive maintenance by exploiting sensor data fusion and artificial intelligence-based analysis. With a dataset such as vibration and sound from sensors, we focus on studying paradigms that orchestrate the most optimal combination of sensors with deep learning sensor fusion algorithms to enable predictive maintenance. In our experimental setup, we used raw data obtained from two sensors, a microphone, and an accelerometer installed on a brushless direct current (BLDC) motor. The data from each sensor were processed individually and, in a second step, merged to create a solid base for analysis. To diagnose BLDC motor faults, this work proposes to use data-level sensor fusion with deep learning methods such as deep convolutional neural networks (DCNNs) for their ability to automatically extract relevant information from the input data, the long short-term memory method (LSTM), and convolutional long short-term memory (CNN-LSTM), a combination of the two previous methods. The results show that in our setup, sound signals outperform vibrations when used individually for training. However, without any feature selection/extraction step, the accuracy of the models improves with data fusion and reaches 98.8%, 93.5%, and 73.6% for the DCNN, CNN-LSTM, and LSTM methods, respectively, 98.8% being a performance that, according to our reading, has never been reached in the analysis of the faults of a BLDC motor without first going through the extraction of the characteristics and their fusion by traditional methods. These results show that it is possible to work with raw data from multiple sensors and achieve good results using deep learning methods without spending time and resources on selecting appropriate features to extract and methods to use for feature extraction and data fusion.

Research on Behavior Monitoring of Elderly Living Alone Based on Wearable Devices and Sensing Technology

Human behavior recognition and status monitoring are current research hotspots, especially in the fields of medical monitoring, smart home, and elderly care. With the development of sensor technology, low-power IC chips, and wireless body sensors, miniature sensor networks can be popularized and applied in daily life. Since the energy consumption of sensor networks is a bottleneck problem that limits its development, this paper designs a multimodal collaborative sensing method for the application scenarios of elderly people living alone to reduce energy consumption in the process of daily behavior perception of the elderly. This method subdivides behavior perception into behavior recognition and status monitoring, determines the optimal sensor combination for identifying monitoring different behaviors based on information theory, and then uses a behavior recognition model modeled by a multiclassifier and a status model modeled by a plurality of two classifiers that are used to perceive user behavior. A large number of experimental results show that compared with the traditional sensor network method, our proposed solution can achieve effective sensing while reducing the energy consumption in the process of data transmission and model calculation, thereby prolonging the working life of the sensing network and realizing long-term and reliable daily behavior perception.

Tool wear monitoring by ensemble learning and sensor fusion using power, sound, vibration, and AE signals

Sensor-fusion and different machine-learning methods were used for tool condition monitoring (TCM) when sawing wood in harsh conditions using power, sound, vibration, and acoustic emission (AE) signals. Tool classification was performed using two ensemble learning (XGBoost and random forest) methods and SVM. It was discussed that the optimal combination of sensors for monitoring is a trade-off between the accuracy of classifiers and the tolerance for sensor redundancy. AE was shown to be the critical sensor, which combined with power signals and XGBoost resulted in ∼92% classification accuracy. Ensemble learning outperformed the SVM and showed superior performance for TCM using multi-sensory-features.

Deep Learning in Gait Parameter Prediction for OA and TKA Patients Wearing IMU Sensors.

Quantitative assessments of patient movement quality in osteoarthritis (OA), specifically spatiotemporal gait parameters (STGPs), can provide in-depth insight into gait patterns, activity types, and changes in mobility after total knee arthroplasty (TKA). A study was conducted to benchmark the ability of multiple deep neural network (DNN) architectures to predict 12 STGPs from inertial measurement unit (IMU) data and to identify an optimal sensor combination, which has yet to be studied for OA and TKA subjects. DNNs were trained using movement data from 29 subjects, walking at slow, normal, and fast paces and evaluated with cross-fold validation over the subjects. Optimal sensor locations were determined by comparing prediction accuracy with 15 IMU configurations (pelvis, thigh, shank, and feet). Percent error across the 12 STGPs ranged from 2.1% (stride time) to 73.7% (toe-out angle) and overall was more accurate in temporal parameters than spatial parameters. The most and least accurate sensor combinations were feet-thighs and singular pelvis, respectively. DNNs showed promising results in predicting STGPs for OA and TKA subjects based on signals from IMU sensors and overcomes the dependency on sensor locations that can hinder the design of patient monitoring systems for clinical application.

A Fusion Method for Satellite Multi-sensor Based on Improved Support Matrix

In this paper, a satellite multi-sensor data fusion method based on improved support matrix is proposed. The cross-correlation function is used to describe the support degree between multiple sensors, and the measured signals are weighted and fused optimally from the SNR perspective. Compared with the method of distance measurement, this method does not need to set threshold and the calculation process is simpler. Simulation results show that this method can objectively reflect the reliability of each sensor, effectively eliminate redundancy and error information, and select the optimal sensor combination.

Design and Evaluation of a Reliable Low-Cost Atmospheric Pollution Station in Urban Environment

The pollution of the air constitutes an environmental risk to health, crops, animals, forests and water. There are several policies for reducing air pollution regarding industry, energy, transportation, and agriculture. Unfortunately, there is limited monitoring of the air quality in cities and rural areas for supervising the accomplishment of these policies. Reliable monitoring of air pollutants is, typically, based on expensive fixed stations, which constitutes a barrier to tackle. This research presents the design, implementation and evaluation of a small, low-cost, station for monitoring atmospheric pollution. The prototype registers ozone ( $O_{3}$ ) and carbon monoxide ( $CO$ ) using inexpensive sensors. To assure high reliability of the measurements obtained by the sensors installed in this station, it is proposed a calibration procedure based on the selection of the best performance analysis of the following machine learning techniques: multiple linear regression, artificial neural networks, and random forest. Additionally, a decision rule is implemented to select an optimal combination of sensors for the estimation models, while the sample timestamp is considered as a temporal heuristic at the input of the system, assuming similarities in the daily environmental dynamics. In order to test the station in a realistic scenario, the calibration and evaluation sets were taken in two different time frames of one and two months, respectively. The overall process was implemented with reference data coming from a certified air quality fixed station in the city of Cuenca - Ecuador. Experimental results showed that the real-time reports of ozone provided by the prototype are quite similar to the fixed station during the evaluation period, with a resulting correlation of up to $r=0.92$ and $r=0.91$ in the calibration and evaluation set, respectively. However, signal drift and aging in $CO_{x}$ sensors diminished the accuracy of carbon monoxide calibration models, resulting in lower correlation ( $r \leq 0.76$ ) with the evaluation set.

Efficient health monitoring of buildings using failure modes and effects analysis case study: Air handling unit system

System health management based on condition monitoring is of great value and significance for improving system performance and reliability. As more data from condition monitoring is obtained, the efficiency of system health monitoring improves. However, increasing the amount of data by installing new sensors on system components is restricted by a variety of factors such as budget, weight and space allowance. Therefore, sensor selection in the most optimal manner in order to achieve maximum valuable information from system is an important issue. In this study, a novel concept is introduced that can be used in sensor combination optimization problems. This concept employs failure modes and effect analysis method in signal information value quantification procedure.To verify the effectiveness of the proposed methodology, an optimization tool is developed based on the proposed methodology for an air handling unit as a case study. This tool offers sensor combination optimization and visualization of key failure mode parameters. Results show that the optimal sensor combination lead to a maximum information on system health considering budget limits. In addition, the sensitivity analysis on severity factor of component failures is performed. Results indicate that, by increasing the severity factor of a component failure, the optimization model select sensors that monitor more information on that failure mode. According to the results obtained, the proposed methodology can be used in maximizing the performance of health monitoring and management of complex systems with numerous applications including aerospace, oil and gas industry, and medical technology.

Unsupervised emotional state classification through physiological parameters for social robotics applications

Future social robots should have personalized behaviors based on user emotional state to fit more in ordinary users’ activities and to improve the human–robot interaction. Several, literature works use cameras to record emotions. However, these approaches may not be effective in everyday life, due to camera obstructions and different types of stimulation, which can be related also with the interaction with other human beings. Therefore, in this work, it is investigated the electrocardiogram, the electrodermal activity, and the electric brain activity physiological signals as main informative channels. The aforementioned signals have been acquired through the use of a wireless wearable sensor network. An experimental methodology was proposed to induce three different emotional states by means of social interaction. Two different combinations of sensors were analyzed using three different time-window frames (180s, 150s, and 120s) and classified with three unsupervised machine learning approaches (K-Means, K-medoids and Self-organizing maps). Finally, their classification performances were compared to the ones obtained by four commonly used supervised techniques (i.e. Support Vector Machine, Decision Tree and k-nearest neighbor) to discuss the optimal combination of sensors, time-window length, and unsupervised classifier. Fifteen healthy young participants were recruited in the study and more than 100 instances were analyzed. The proposed approaches achieve an accuracy of 77% in the best-unsupervised case and 85% with the best-supervised ones.

Optimal linear combination of sensors and actuators to enforce an interior conic open‐loop system

SummaryThis paper presents techniques to linearly combine the sensor measurements and/or actuator inputs of a linear time‐invariant system to obtain a new system that is interior conic with prescribed bounds. In the optimal sensor combination problem, a desired system output is defined, and in the optimal actuator combination problem, a desired system input is defined, along with a frequency bandwidth in which the desired system input or output should be matched. The simultaneous optimal sensor and actuator combination problem includes desired system outputs and inputs. In all cases, the weighted or norm of the difference between the system with linearly combined sensors or actuators and the desired system is minimized while rendering the new system interior conic with prescribed bounds. The weighting transfer matrix used in the ‐ or ‐optimization problem is determined by the frequency bandwidth of interest. The individual sensor and actuator combination methods involve linear matrix inequality constraints and are posed as convex optimization problems, whereas the combined sensor and actuator method is an iterative procedure composed of convex optimization steps. Numerical examples illustrate superior tracking performance with the proposed sensor and actuator combination techniques over comparable techniques in the literature when implemented with a simple feedback controller. Robust asymptotic stability of the closed‐loop system to plant uncertainty is demonstrated in the numerical examples.

Innovative sprayer with optimized sensor-based gap detection for precision application of plant protection products

For fruit growing, plant protection products (PPP) are required to maintain yield and to fulfill high quality standards. In vertical crops the application of PPP is made by active air support, which can result in unwanted displacement of PPP into the environment. These amounts rise when trees are bare and gaps in the canopy are larger. In densely populated regions and those with a high density of ditches, the reduction of these amounts and drift mitigation are important sociopolitical goals. Modern sensor systems offer the possibility for precise target and gap detection and can ensure an appropriate switching of single nozzles. This innovative system allows a high saving potential of PPP and a precise application onto the target area in each tree. In the present project, a concept of an optimal combination of sensors and nozzles for gap detection was further developed. Different sprayers were equipped with a high number of new optical infrared-sensors to optimize the scanning field in the orchard. In comparison with previous sensor systems, the new sensors detect single bare branches without focusing a second tree row and neglect branches in the immediate area. In 2014, field experiments in Jork have shown that a precise application can reduce the output quantity of PPP, as much as 40% in a ten-year-old apple orchard. The efficiency was significantly influenced by software-supported switch-on and switch-off of the nozzles. The evaluation of the losses in the gap showed lower values compared to conventional full application, which could also result in a drift reduction and more efficient application onto the target area. The aim of the project is to develop marketable sprayers with different fan types as well as a retrofitting kit for sprayers in use, which fulfill the requirements for a high PPP usage efficiency.

Simultaneous measurement of multiple soil properties through proximal sensor data fusion: A case study

In this research, proximal soil sensor data fusion was defined as a multifaceted process which integrates geospatially correlated data, or information, from multiple proximal soil sensors to accurately characterize the spatial complexity of soils. This has capability of providing improved understanding of soil heterogeneity for potential applications associated with crop production and natural resource management. To assess the potential of data fusion for the purpose of improving thematic soil mapping over the single sensor approach, data from multiple proximal soil sensors were combined to develop and validate predictive relationships with laboratory-measured soil physical and chemical properties. The work was conducted in an agricultural field with both mineral and organic soils. The integrated data included: topography records obtained using a real-time kinetic (RTK) global navigation satellite system (GNSS) receiver, apparent soil electrical conductivity (ECa) obtained using an electromagnetic induction sensor, and content of several naturally occurring radioisotopes detected using a mobile gamma-ray spectrometer. In addition, the soil profile data were collected using a commercial ruggedized multi-sensor platform carrying a visible and near-infrared (vis-NIR) optical sensor and a galvanic contact soil ECa sensor. The measurements were carried out at predefined field locations covering the entire study area identified from sensor measured a priori information on field elevation, ECa and gamma-ray count. The information was used to predict: soil organic matter (SOM), pH, lime buffer capacity (LBC), as well as concentration of phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and aluminum (Al). Partial least squares regressions (PLSRs) were used to predict soil properties from individual sensors and different sensor combinations (sensor data fusion). By integrating the data from all of the proximal soil sensors, SOM, pH, LBC, Ca, Mg, and Al were predicted simultaneously with R2 > 0.5 (RPD > 1.50). Improved predictions were observed for most soil properties based on sensor data fusion than those based on individual sensors. After choosing the optimal sensor combination for each soil property, the predictive capability was compared using different data mining algorithms, including support vector machines (SVM), random forest (RF), multivariate adaptive regression splines (MARS), and regression trees (CART). Improved predictions for SOM, Ca, Mg, and Al were observed using SVM over PLSR. The predictive capability was followed by RF and MARS, with CART. Predictions of pH and LBC were only feasible using MARS and PLSR, respectively. In this field, it was not possible to predict extractable P and K using all tested sensor combinations or algorithms. With large variability in SOM, the field presents a special situation and thus, the result could be specific to the study site. Further research includes an extended number of experimental sites covering different geographic areas around Eastern Canada.

Node Selection Algorithm for Underwater Acoustic Sensor Network Based on Particle Swarm Optimization

Underwater target positioning technology is the most important part of UnderWater Acoustic Sensor Network(called UWASN), and it is one of the most important research directions in this field with broad application prospects in commercial and military fields. Due to the complex and variability of underwater acoustic environment, the underwater acoustic sensor network has the characteristics of fluidity, sparse deployment and energy limitation, which brings certain challenges to underwater positioning technology. Aiming at the scenario that the node redundancy in the underwater acoustic sensor network leads to low positioning efficiency, this paper considers the sound velocity correction factor based on the traditional anchor node selection algorithm in this paper. Under the premise of ensuring certain positioning accuracy, considering the communication overhead, node residual energy, position suspiciousness, sound ray propagation bending characteristics and other factors, the anchor node optimization mechanism which uses the particle swarm algorithm to iterate out the optimal sensor combination for improving the accuracy of positioning is designed. The simulation results show that the proposed algorithm shows small calculation, fast convergence and high positioning accuracy. It can effectively improve the energy utilization of nodes, balance positioning performance as well as energy use efficiency, and optimize the positioning result of UWASN, which is well suited for underwater acoustic positioning scenarios.

Hierarchical Analysis of Thorax Models to Measure Tidal Volume

Abstract Motion tracking of thorax kinematics can be used to determine respiration. However, determining a minimal sensor configuration from 64 candidate sensor locations is associated with high computational costs. Hence, a hierarchical optimization method was proposed to determine the optimal combination of sensors. The hierarchical method was assessed by its ability to quickly determine the sensor combination that will yield optimal modelled tidal volume compared to body plethysmograph measurements. This method was able to find the optimal sensor combinations, in approximately 2% of the estimated time required by an exhaustive search.

Motion Artifact Quantification and Sensor Fusion for Unobtrusive Health Monitoring.

Sensors integrated into objects of everyday life potentially allow unobtrusive health monitoring at home. However, since the coupling of sensors and subject is not as well-defined as compared to a clinical setting, the signal quality is much more variable and can be disturbed significantly by motion artifacts. One way of tackling this challenge is the combined evaluation of multiple channels via sensor fusion. For robust and accurate sensor fusion, analyzing the influence of motion on different modalities is crucial. In this work, a multimodal sensor setup integrated into an armchair is presented that combines capacitively coupled electrocardiography, reflective photoplethysmography, two high-frequency impedance sensors and two types of ballistocardiography sensors. To quantify motion artifacts, a motion protocol performed by healthy volunteers is recorded with a motion capture system, and reference sensors perform cardiorespiratory monitoring. The shape-based signal-to-noise ratio SNR is introduced and used to quantify the effect on motion on different sensing modalities. Based on this analysis, an optimal combination of sensors and fusion methodology is developed and evaluated. Using the proposed approach, beat-to-beat heart-rate is estimated with a coverage of 99.5% and a mean absolute error of 7.9 ms on 425 min of data from seven volunteers in a proof-of-concept measurement scenario.

Multi-modal myocontrol: Testing combined force- and electromyography.

Myocontrol, that is control of prostheses using bodily signals, has proved in the decades to be a surprisingly hard problem for the scientific community of assistive and rehabilitation robotics. In particular, traditional surface electromyography (sEMG) seems to be no longer enough to guarantee dexterity (i.e., control over several degrees of freedom) and, most importantly, reliability. Multi-modal myocontrol is concerned with the idea of using novel signal gathering techniques as a replacement of, or alongside, sEMG, to provide high-density and diverse signals to improve dexterity and make the control more reliable. In this paper we present an offline and online assessment of multi-modal sEMG and force myography (FMG) targeted at hand and wrist myocontrol. A total number of twenty sEMG and FMG sensors were used simultaneously, in several combined configurations, to predict opening/closing of the hand and activation of two degrees of freedom of the wrist of ten intact subjects. The analysis was targeted at determining the optimal sensor combination and control parameters; the experimental results indicate that sEMG sensors alone perform worst, yielding a nRMSE of 9.1%, while mixing FMG and sEMG or using FMG only reduces the nRMSE to 5.2-6.6%. To validate these results, we engaged the subject with median performance in an online goal-reaching task. Analysis of this further experiment reveals that the online behaviour is similar to the offline one.

Multivariate respiratory motion prediction

In extracranial robotic radiotherapy, tumour motion is compensated by tracking external and internal surrogates. To compensate system specific time delays, time series prediction of the external optical surrogates is used. We investigate whether the prediction accuracy can be increased by expanding the current clinical setup by an accelerometer, a strain belt and a flow sensor. Four previously published prediction algorithms are adapted to multivariate inputs—normalized least mean squares (nLMS), wavelet-based least mean squares (wLMS), support vector regression (SVR) and relevance vector machines (RVM)—and evaluated for three different prediction horizons. The measurement involves 18 subjects and consists of two phases, focusing on long term trends (M1) and breathing artefacts (M2). To select the most relevant and least redundant sensors, a sequential forward selection (SFS) method is proposed. Using a multivariate setting, the results show that the clinically used nLMS algorithm is susceptible to large outliers. In the case of irregular breathing (M2), the mean root mean square error (RMSE) of a univariate nLMS algorithm is 0.66 mm and can be decreased to 0.46 mm by a multivariate RVM model (best algorithm on average). To investigate the full potential of this approach, the optimal sensor combination was also estimated on the complete test set. The results indicate that a further decrease in RMSE is possible for RVM (to 0.42 mm). This motivates further research about sensor selection methods. Besides the optical surrogates, the sensors most frequently selected by the algorithms are the accelerometer and the strain belt. These sensors could be easily integrated in the current clinical setup and would allow a more precise motion compensation.