Explaining prediction models and individual predictions with feature contributions

We present a sensitivity analysis-based method for explaining prediction models that can be applied to any type of classification or regression model. Its advantage over existing general methods is that all subsets of input features are perturbed, so interactions and redundancies between features are taken into account. Furthermore, when explaining an additive model, the method is equivalent to commonly used additive model-specific methods. We illustrate the method’s usefulness with examples from artificial and real-world data sets and an empirical analysis of running times. Results from a controlled experiment with 122 participants suggest that the method’s explanations improved the participants’ understanding of the model.     

Control of mineral wool thickness using predictive functional control

The production process of mineral wool is affected by several constantly changing factors. The ingredients for the mineral wool are melted in a furnace. The molten mineral charge exits the bottom of the furnace in a water-cooled trough and falls into a fiberization device (the centrifuge). The centrifuge forms the fibers. At this stage binders are injected to bind the fibers together. To ensure the quality of the end product (the consistent thickness) the flow of the bounded fibers must be as constant as possible. One way to ensure that is to control the speed of the conveyor belt that transports the bounded fibers from the centrifuge to the curing process. Predictive functional controller and PID controller are considered to replace an existing algorithm. Both can easily replace an existing one as they do not require any new sensor installation. All three algorithms are presented and tested on a developed plant model. The study showed that the predictive control gives better results than the existing and PID controller.     

Charge Transfer Variability in Misfit Layer Compounds: Comparison of SnS-SnS2 and LaS-TaS2

The present paper compares and discusses two selected misfit (layer) compounds exemplarily, namely SnS-SnS₂ and LaS-TaS₂. We have employed a density-functional theory-based approach to calculate structural, energetic, and electronic properties of these structures. We have put emphasis on the difference between single layers, combined double-layer systems and periodically stacked bulk structures. Especially the varying magnitudes of charge transfer between the sublayers were studied. We demonstrate how the chemical constitution of the sublayers affects the interlayer interactions: these may be a weak non-bonding van-der-Waals dominated interlayer interaction as in SnS-SnS₂ and many other layered structures or a strong interaction related to a remarkable charge transfer between the layers as in LaS-TaS₂.     

Design of a Telescopic Linear Actuator Based on Hollow Shape Memory Springs

Shape memory alloys (SMAs) are smart materials exploited in many applications to build actuators with high power to mass ratio. Typical SMA drawbacks are: wires show poor stroke and excessive length, helical springs have limited mechanical bandwidth and high power consumption. This study is focused on the design of a large-scale linear SMA actuator conceived to maximize the stroke while limiting the overall size and the electric consumption. This result is achieved by adopting for the actuator a telescopic multi-stage architecture and using SMA helical springs with hollow cross section to power the stages. The hollow geometry leads to reduced axial size and mass of the actuator and to enhanced working frequency while the telescopic design confers to the actuator an indexable motion, with a number of different displacements being achieved through simple on-off control strategies. An analytical thermo-electro-mechanical model is developed to optimize the device. Output stroke and force are maximized while total size and power consumption are simultaneously minimized. Finally, the optimized actuator, showing good performance from all these points of view, is designed in detail.     

A novel clipping technique for filtering FM signals embedded in intensive noise

A novel adaptive clipping technique for filtering a constant amplitude frequency modulated (FM) signal embedded in non-Gaussian noise is proposed. It is based on the analysis and processing of the estimate of probability density function of a FM signal realization. As a result, modifications of two robust estimators of FM signal amplitude are proposed. It is shown that these estimators can be used for Gaussian and non-Gaussian heavy-tail environments. The proposed clipping technique can exploit one or another obtained robust estimate of the signal amplitude for adaptive setting a threshold. Analysis of signal estimate accuracy for different noise environments is carried out. Comparative analysis of the obtained methods and known approaches based on scanning window nonlinear filtering and optimal robust L-DFT form is performed. It is demonstrated that the usage of clipping-based technique leads to the considerable improvement of the FM signal filtering efficiency in comparison to the aforementioned known approaches for different noise environments and a wide range of input SNR values.     

Surfactant-Assisted Dispersion of MWCNTs in Epoxy Resin Used in CFRP Strengthening Systems

The epoxy resin used as the bonding agent in carbon fiber-reinforced polymer (CFRP) strengthening systems was modified by the infusion of multiwalled carbon nanotubes (MWCNTs). Two types of surfactants, Triton X-100 and C₁₂E₈, were used to disperse the nanotubes in the epoxy resin employing ultrasonic mixing. Dynamic mechanical analysis and tensile tests were conducted to study the effect of the surfactant-assisted dispersion of nanotubes on the thermal and mechanical properties of epoxy composites. The morphology of the epoxy composites was interpreted using scanning electron microscopy (SEM). Moreover, the effect of surfactant treatment on the structure of nanotubes was investigated by Fourier transform infrared (FT-IR). Based on the experimental results, the tensile strength and the storage modulus of the epoxy resin were increased by 32% and 26%, respectively, by the addition of MWCNTs. This was attributed to the homogeneous dispersion of nanotubes in the epoxy resin according to the SEM images. Another reason for the enhancement in the tensile properties was the reinforced nanotube/epoxy interaction as a result of the surfactant anchoring effect which was proved by FT-IR. A moderate improvement in the glass transition temperature (T _g) was recorded for the composite fabricated using Triton X-100, which was due to the restricted molecular motions in the epoxy matrix. To characterize the temperature-dependent tensile behavior of the modified epoxy composites, tensile tests were conducted at elevated temperatures. It was revealed that the MWCNT modification using surfactant substantially improves the tensile performance of the epoxy adhesive at temperatures above the T _g of the neat epoxy.