An end-to-end machine learning approach with explanation for time series with varying lengths. - In: Neural computing & applications, ISSN 1433-3058, Bd. 36 (2024), 13, S. 7491-7508
An accurate prediction of complex product quality parameters from process time series by an end-to-end learning approach remains a significant challenge in machine learning. A special difficulty is the application of industrial batch process data because many batch processes generate variable length time series. In the industrial application of such methods, explainability is often desired. In this study, a 1D convolutional neural network (CNN) algorithm with a masking layer is proposed to solve the problem for time series of variable length. In addition, a novel combination of 1D CNN and class activation mapping (CAM) technique is part of this study to better understand the model results and highlight some regions of interest in the time series. As a comparative state-of-the-art unsupervised machine learning method, the One-Nearest Neighbours (1NN) algorithm combined with dynamic time warping (DTW) was used. Both methods are investigated as end-to-end learning methods with balanced and unbalanced class distributions and with scaled and unscaled input data, respectively. The FastDTW and DTAIDistance algorithms were investigated for the DTW calculation. The data set is made up of sensor signals that was collected during the production of plastic parts. The objective was to predict a quality parameter of plastic parts during production. For this research, the quality parameter will be a difficult or only destructively measurable parameter and both methods will be investigated for their applicability to this prediction task. The application of the proposed approach to an industrial facility for producing plastic products shows a prediction accuracy of 83.7%. It can improve the reverence method by approximately 1.4%. In addition to the slight increase in accuracy, the CNN training time was significantly reduced compared to the DTW calculation.
https://doi.org/10.1007/s00521-024-09473-9
Modified implicit discretization of the super-twisting controller. - In: IEEE transactions on automatic control, ISSN 1558-2523, Bd. 0 (2024), 0, S. 1-8
In this paper a novel discrete-time realization of the super-twisting controller is proposed. The closed-loop system is proven to converge to an invariant set around the origin in finite time. Furthermore, the steady-state error is shown to be independent of the controller gains. It only depends on the sampling time and the unknown disturbance. The proposed discrete-time controller is evaluated comparative to previously published discrete-time super-twisting controllers by means of the controller structure and in extensive simulation studies. The continuous-time super-twisting controller is capable of rejecting any unknown Lipschitz-continuous perturbation and converges in finite time. Furthermore, the convergence time decreases, if any of the gains is increased. The simulations demonstrate that the closed-loop systems with each of the known controllers lose one of these properties, introduce discretization-chattering, or do not yield the same accuracy level as with the proposed controller. The proposed controller, in contrast, is beneficial in terms of the above described properties.
https://doi.org/10.1109/TAC.2024.3370494
Modal logic is more succinct iff bi-implication is available in some form. - In: 41st International Symposium on Theoretical Aspects of Computer Science, (2024), S. 12:1-12:17
Is it possible to write significantly smaller formulae, when using more Boolean operators in addition to the De Morgan basis (and, or, not)? For propositional logic a negative answer was given by Pratt: every formula with additional operators can be translated to the De Morgan basis with only polynomial increase in size. Surprisingly, for modal logic the picture is different: we show that adding bi-implication allows to write exponentially smaller formulae. Moreover, we provide a complete classification of finite sets of Boolean operators showing they are either of no help (allow polynomial translations to the De Morgan basis) or can express properties as succinct as modal logic with additional bi-implication. More precisely, these results are shown for the modal logic T (and therefore for K). We complement this result showing that the modal logic S5 behaves as propositional logic: no additional Boolean operators make it possible to write significantly smaller formulae.
https://doi.org/10.4230/LIPIcs.STACS.2024.12
A characterization of efficiently compilable constraint languages. - In: 41st International Symposium on Theoretical Aspects of Computer Science, (2024), S. 11:1-11:19
A central task in knowledge compilation is to compile a CNF-SAT instance into a succinct representation format that allows efficient operations such as testing satisfiability, counting, or enumerating all solutions. Useful representation formats studied in this area range from ordered binary decision diagrams (OBDDs) to circuits in decomposable negation normal form (DNNFs). While it is known that there exist CNF formulas that require exponential size representations, the situation is less well studied for other types of constraints than Boolean disjunctive clauses. The constraint satisfaction problem (CSP) is a powerful framework that generalizes CNF-SAT by allowing arbitrary sets of constraints over any finite domain. The main goal of our work is to understand for which type of constraints (also called the constraint language) it is possible to efficiently compute representations of polynomial size. We answer this question completely and prove two tight characterizations of efficiently compilable constraint languages, depending on whether target format is structured. We first identify the combinatorial property of "strong blockwise decomposability" and show that if a constraint language has this property, we can compute DNNF representations of linear size. For all other constraint languages we construct families of CSP-instances that provably require DNNFs of exponential size. For a subclass of "strong uniformly blockwise decomposable" constraint languages we obtain a similar dichotomy for structured DNNFs. In fact, strong (uniform) blockwise decomposability even allows efficient compilation into multi-valued analogs of OBDDs and FBDDs, respectively. Thus, we get complete characterizations for all knowledge compilation classes between O(B)DDs and DNNFs.
https://doi.org/10.4230/LIPIcs.STACS.2024.11
Computational flow diverter implantation - a comparative study on pre-interventional simulation and post-interventional device positioning for a novel blood flow modulator. - In: Fluids, ISSN 2311-5521, Bd. 9 (2024), 3, 55, S. 1-15
Due to their effect on aneurysm hemodynamics, flow diverters (FD) have become a routine endovascular therapy for intracranial aneurysms. Since over- and undersizing affect the device’s hemodynamic abilities, selecting the correct device diameter and accurately simulating FD placement can improve patient-specific outcomes. The purpose of this study was to validate the accuracy of virtual flow diverter deployments in the novel Derivo® 2 device. We retrospectively analyzed blood flows in ten FD placements for which 3D DSA datasets were available pre- and post-intervention. All patients were treated with a second-generation FD Derivo® 2 (Acandis GmbH, Pforzheim, Germany) and post-interventional datasets were compared to virtual FD deployment at the implanted position for implanted stent length, stent diameters, and curvature analysis using ANKYRAS (Galgo Medical, Barcelona, Spain). Image-based blood flow simulations of pre- and post-interventional configurations were conducted. The mean length of implanted FD was 32.61 (±11.18 mm). Overall, ANKYRAS prediction was good with an average deviation of 8.4% (±5.8%) with a mean absolute difference in stent length of 3.13 mm. There was a difference of 0.24 mm in stent diameter amplitude toward ANKYRAS simulation. In vessels exhibiting a high degree of curvature, however, relevant differences between simulated and real-patient data were observed. The intrasaccular blood flow activity represented by the wall shear stress was qualitatively reduced in all cases. Inflow velocity decreased and the pulsatility over the cardiac cycle was weakened. Virtual stenting is an accurate tool for FD positioning, which may help facilitate flow FDs’ individualization and assess their hemodynamic impact. Challenges posed by complex vessel anatomy and high curvatures must be addressed.
https://doi.org/10.3390/fluids9030055
Simultaneous minimization of water age and pressure in water distribution systems by pressure reducing valves. - In: Water resources management, ISSN 1573-1650, Bd. 0 (2024), 0, insges. 19 S.
Pressure reducing valves (PRVs) are essentially used to reduce operational pressures in water distribution systems (WDSs) to minimize water leakage. However, water age in a WDS is an important variable describing the water quality and should be kept as low as possible. Therefore, the aim of this study is to investigate the possibility and potential of simultaneously minimizing both pressure and water age by using PRVs. To determine the optimal location and setting of PRVs, a mixed-integer nonlinear programming (MINLP) problem is formulated with minimization of the sum of the weighted total water age and pressure as the objective function, where the weighting factor can be defined by the user’s preference. The equality constraints consist of the hydraulic equations and water age functions to describe pressure and water age in the distribution network, while the inequality constraints ensure them in the defined operating ranges, respectively. Applying the proposed approach to two case studies, the results show that both water age and pressure can indeed be significantly reduced by the optimized position and setting of the PRVs.
https://doi.org/10.1007/s11269-024-03828-6
Dynamic-controlled Bayesian network for process pattern modeling and optimization. - In: Industrial & engineering chemistry research, ISSN 1520-5045, Bd. 0 (2024), 0, S. 1-11
Capturing the current statistical features of a process and its dynamic evolution is important for controlling and monitoring its overall operational status. In terms of capturing the process dynamics, existing probabilistic latent-variable methods mostly consider autoregressive relationships, and thus, the causality from the control inputs to the pattern, or key hidden variable, remains unmodeled or implicit. To bridge this gap, a model structured by a newly designed dynamic-controlled Bayesian network (DCBN) is proposed in this paper for pattern modeling, especially pattern control and optimization. Significantly, the innovation and advantage of the DCBN lie in explicitly quantifying the impulse response of the pattern under control inputs. As well, the expectation-maximization algorithm is specially designed for learning the DCBN model. Finally, a new framework for pattern-based process control and optimization is presented in which online pattern filtering and control can be implemented. A case study on the combustion process from an industrial boiler illustrates the advantages of the proposed method in that it can capture the controlled dynamics of the process and achieve optimization by tracking the pattern set point or trajectory.
https://doi.org/10.1021/acs.iecr.3c04391
Advanced deep learning for skin histoglyphics at cellular level. - In: Bildverarbeitung für die Medizin 2024, (2024), S. 66-71
In dermatology, the histological examination of skin cross-sections is essential for skin cancer diagnosis and treatment planning. However, the complete coverage of tissue abnormalities is not possible due to time constraints as well as the sheer number of cell groups. We present an automatic segmentation approach of seven tissue classes: vessels, perspiration glands, hair follicles, sebaceous glands, tumor tissue, epidermis and fatty tissue, for a fast processing of the large datasets. Hence, the initial size of the data lends itself to the use of patch-based deep learning models, resulting in good IoU score of 94.2 percent for the cancerous tissue and overall IoU score of 83.6 percent.
https://doi.org/10.1007/978-3-658-44037-4_20
Targeted excitation and re-identification methods for multivariate process and model predictive control. - In: Journal of process control, ISSN 0959-1524, Bd. 136 (2024), 103190, S. 1-15
A process controlled using model predictive control is required to be re-identified when significant plant-model mismatch (PMM) occurs. During data acquisition for re-identification, the process is excited to enable accurate re-identification. However, the process excitation worsens the control performance. To prevent this problem, a new model-update framework that consists of targeted excitation (TE) and targeted re-identification (TR) is proposed. In TE, only the manipulated variables corresponding to problematic transfer functions that have significant PMM are excited during data acquisition. On the other hand, the other manipulated variables are optimized to suppress the variations of the controlled variables. After data is acquired using TE, the TR method re-identifies only the problematic transfer functions by using the other transfer-function models without large PMM. The validity of the proposed framework is examined by theoretical analysis and numerical case studies. In the theoretical analysis, the stability during data acquisition using TE and the asymptotic bias of the parameters re-identified using TR were considered. In the numerical case studies, the applicability of the proposed framework to several processes including a fluid catalytic cracking (FCC) process was examined. As a result, it was shown that, for all the processes, the proposed framework can improve both the control performance during data acquisition and the model accuracy after re-identification, compared to an existing method that excites all the inputs during data acquisition.
https://doi.org/10.1016/j.jprocont.2024.103190
Spectral separation of evoked and spontaneous cortical activity, Part 1: Delta to high gamma band. - In: Biomedical signal processing and control, ISSN 1746-8108, Bd. 92 (2024), 106094, S. 1-11
Spectral analysis of repeatedly evoked potentials (EPs) is challenging since recordings contain a superposition of evoked signals and spontaneous activity. We developed a novel approach, N-interval Fourier Transform Analysis (N-FTA), which allows for reliable separation and simultaneous assessment of triggered and background spectral components. Median nerve stimulation data from a total of eleven volunteers recorded in two labs with different experimental settings were investigated. Consistently, short latency spectral components were mainly contained in the gamma and high gamma bands. In contrast, spontaneous activity displayed a 1/f spectral profile with distinct alpha and beta peaks. Spontaneous power spectral densities (PSDs) obtained for real and sham stimulation were highly comparable. The low frequency background PSD was more than two orders of magnitude above the spectral short latency peaks. Within the 30Hz to 90Hz band, the evoked peaks were -17dB to -4dB below the background suggesting that target band filtered short latency deflection might be extracted using less than 100 trials. SEPs following tibial nerve stimulation (3 subjects) displayed a narrower spectral band at about half the bandwidth as compared to median nerve stimulation. Evoked peaks were between 30Hz and 37Hz at PSD levels being -10dB to -4dB below the background activity. These spectral peaks were related to the short latency response of typical W-morphology. Cortical short latency responses are contained in distinct spectral target bands which are much narrower than the standard bandwidth recommendations for routine recordings. In particular, the high pass corner frequency may be selected about one order of magnitude above the current standard. This might render SEP recordings more robust since it eases the suppression of spontaneous activity and movement artifacts such as eye-blinks. Real-time zero-phase filters are required for translating these findings into improved recording systems.
https://doi.org/10.1016/j.bspc.2024.106094