Soft Sensor Development Research Articles

Wireless batteryless soft sensors for ambulatory cardiovascular health monitoring

Seismocardiography (SCG) is the measure of local vibrations in the chest due to heartbeats. Typically, SCG signals are measured using rigid integrated circuit (IC) accelerometers and bulky electronics. However, as alternatives, recent studies of emerging flexible sensors show promise. Here, we introduce the development of wireless soft capacitive sensors that require no battery or rigid IC components for measuring SCG signals for cardiovascular health monitoring. Both the capacitive and inductive components of the circuit are patterned with laser micromachining of a polyimide-coated copper and are encapsulated with an elastomer. The wearable soft sensor can detect small strain changes on the skin, which is wirelessly measured by examining the power reflected from the antenna at a stimulating frequency. The performance of the device is verified by comparing the fiducial points to SCG measured by a commercial accelerometer and electromyograms from a commercial electrode. Overall, the human subject study demonstrates that the fiducial points are consistent with data from commercial devices, showing the potential of the ultrathin soft sensors for ambulatory cardiovascular monitoring without bulky electronics and rigid components.

Data Quality Over Quantity: Pitfalls and Guidelines for Process Analytics

A significant portion of the effort involved in advanced process control, process analytics, and machine learning involves acquiring and preparing data. Literature often emphasizes increasingly complex modelling techniques with incremental performance improvements. However, when industrial case studies are published they often lack important details on data acquisition and preparation. Although data pre-processing is unfairly maligned as trivial and technically uninteresting, in practice it has an out-sized influence on the success of real-world Artificial intelligence applications. This work describes best practices for acquiring and preparing operating data to pursue data-driven modelling and control opportunities in industrial processes. We present practical considerations for pre-processing industrial time series data to inform the efficient development of reliable soft sensors that provide valuable process insights.

Highly Stretchable Additively Manufactured Capacitive Proximity and Tactile Sensors for Soft Robotic Systems

Soft robotics are considered one of the most promising approaches towards fully collaborative robotic devices. Soft robotic systems are intrinsically safe due to their compliant nature. Using additive fabrication for the development of soft capacitive sensors, enables high flexibility and individualization capabilities in terms of design and material properties. In this work, we outline the sensorization of stretchable capacitive sensors for proximity and tactile detection. This sensor will represent an essential component of a soft robotic finger in future work. While the functional principle has already been shown previously, here we explore the design space further and perform more in depth test of the influences of the strain on the tactile measurements. The capacitive pad’s sensitivity is evaluated by applying normal force on each sensor, where the applied force is ranged between 1N and 16N, with resultant capacitive change of the four tactile sensors between 1 pF and 60 pF. The capacitive pad’s proximity sensor is tested using five different materials, with proximity range of 20mm, with change in capacitance between 0.1 pF and 2.4 pF. An essential aspect of soft robotics is that also the sensor may be subject to substantial deformations. We demonstrate that our proposed sensor design can withstand large strains of more than 35% while the sensitivity towards sensing normal forces does not significantly change. This demonstrates that the proposed sensor design will be suitable for various soft robotics applications, where large deformations may occur.

The importance of outlier rejection and significant explanatory variable selection for pinot noir wine soft sensor development

Sensory attributes are essential factors in determining the quality of wines. However, it can be challenging for consumers, even experts, to differentiate and quantify wines' sensory attributes for quality control. Soft sensors based on rapid chemical analysis offer a potential solution to overcome this challenge. However, the current limitation in developing soft sensors for wines is the need for a significant number of input parameters, at least 12, necessitating costly and time-consuming analyses. While such a comprehensive approach provides high accuracy in sensory quality mapping, the expensive and time-consuming studies required do not lend themselves to the industry's routine quality control activities. In this work, Box plots, Tucker-1 plots, and Principal Component Analysis (PCA) score plots were used to deal with output data (sensory attributes) to improve the model quality. More importantly, this work has identified that the number of analyses required to fully quantify by regression models and qualify by classification models can be significantly reduced. Based on regression models, only four key chemical parameters (total flavanols, total tannins, A520nmHCl, and pH) were required to accurately predict 35 sensory attributes of a wine with R2 values above 0.6 simultaneously. In addition, for classification models to accurately predict 35 sensory attributes of a wine at once with prediction accuracy above 70%, only four key chemical parameters (A280nmHCl, A520nmHCl, chemical age and pH) were required. These models with reduced chemical parameters complement each other in sensory quality mapping and provide acceptable accuracy. The application of the soft sensor based on these reduced sets of key chemical parameters translated to a potential reduction in analytical cost and labour cost of 56% for the regression model and 83% for the classification model, respectively, making these models suitable for routine quality control use.

Soft-Sensor Development through Deep Learning with Spatial and Temporal Feature Extraction of Complex Processes

Data-driven soft sensing techniques have been widely applied to predict the quality variables in the process industry. The temporal and spatial structure information is important for process modeling because it helps extract the nonlinear and dynamic characteristics from data. Deep learning can be used to mimic the physical linking structure by designing similar node connections in the neural network. Taking a complex process with countercurrent unit connection as a demonstration, this paper proposes using a bidirectional long short-term memory-based network with spatiotemporal feature extraction for soft-sensor development. An attention mechanism is further conducted for the feature learning from space and time simultaneously. The effectiveness of the proposed method is validated through a simulation example and a real industrial example.

Profitability related industrial-scale batch processes monitoring via deep learning based soft sensor development

Data-driven soft sensor technology has been widely developed to estimate quality-related variables, while following difficulties still limit its application in batch processes, such as different initial conditions, uneven-length of batches, and the extraction of within-batch multiphase features. To address these problems, a quality prediction and monitoring framework is proposed. Variables related to quality-related variables are first selected, and a data stacked strategy is proposed to transform three-dimensional batch data into time-lagged sequences that can be fed into soft sensor models. Aiming to extract the multiphase features, a novel differential recurrent neural networks is established by embedding differential operations into long short-term memory neural networks. Moreover, to ensure profitability, prediction residuals are employed for quality monitoring. Case study on a simulation dataset and an industrial-scale penicillin fermentation process demonstrates the effectiveness of the proposed method and its applicability to batch process monitoring and control in both academic research and industrial operation.

Variational Progressive-Transfer Network for Soft Sensing of Multirate Industrial Processes

Deep-learning-based soft sensors have been extensively developed for predicting key quality or performance variables in industrial processes. However, most approaches assume that data are uniformly sampled while the multiple variables are often acquired at different rates in practical processes. This article designed a progressive transfer strategy, based on which a variational progressive-transfer network (VPTN) method is proposed for the soft sensor development of industrial multirate processes. In VPTN, the multirate data are first separated into multiple data chunks where the variables within each chunk are acquired at a uniform rate. Then, a variational multichunk data modeling framework is developed to model the multiple chunks in a unified fashion through deep variational structures. The base models, including the unsupervised ones with only partial process variables and the supervised soft sensor model share a similar network structure, such that the subsequent transfer strategy can be readily implemented. Finally, a progressive transfer learning strategy is designed to transfer the model parameters from the fastest sampled data chunk to the slowest one in a progressive manner. Thus, the knowledge from various data chunks can be sequentially explored and transferred to enhance the performance of the terminal soft sensor model. Case studies on both a debutanizer column dataset and a real coal mill dataset in a thermal power plant validate the performance of the proposed method.

A new 3D, microfluidic-oriented, multi-functional, and highly stretchable soft wearable sensor

Increasing demand for wearable devices has resulted in the development of soft sensors; however, an excellent soft sensor for measuring stretch, twist, and pressure simultaneously has not been proposed yet. This paper presents a novel, fully 3D, microfluidic-oriented, gel-based, and highly stretchable resistive soft sensor. The proposed sensor is multi-functional and could be used to measure stretch, twist, and pressure, which is the potential of using a fully 3D structure in the sensor. Unlike previous methods, in which almost all of them used EGaIn as the conductive material, in this case, we used a low-cost, safe (biocompatible), and ubiquitous conductive gel instead. To show the functionality of the proposed sensor, FEM simulations and a set of designed experiments were done, which show linear (99%), accurate (> 94.9%), and durable (tested for a whole of four hours) response of the proposed sensor. Then, the sensor was put through its paces on a female test subject’s knee, elbow, and wrist to show the potential application of the sensor as a body motion sensor. Also, a fully 3D active foot insole was developed, fabricated, and evaluated to evaluate the pressure functionality of the sensor. The result shows good discrimination and pressure measurement for different foot sole areas. The proposed sensor has the potential to be used in real-world applications like rehabilitation, wearable devices, soft robotics, smart clothing, gait analysis, AR/VR, etc.

Development of a Robust Data-Driven Soft Sensor for Multivariate Industrial Processes with Non-Gaussian Noise and Outliers

Industrial processes are often nonlinear and multivariate and suffer from non-Gaussian noise and outliers in the process data, which cause significant challenges in data-driven modelling. To address these issues, a robust soft-sensing algorithm that integrates Huber’s M-estimation and adaptive regularisations with multilayer perceptron (MLP) is proposed in this paper. The proposed algorithm, called RAdLASSO-MLP, starts with an initially well-trained MLP for nonlinear data-driven modelling. Subsequently, the residuals of the proposed model are robustified with Huber’s M-estimation to improve the resistance to non-Gaussian noise and outliers. Moreover, a double L1-regularisation mechanism is introduced to minimise redundancies in the input and hidden layers of MLP. In addition, the maximal information coefficient (MIC) index is investigated and used to design the adaptive operator for the L1-regularisation of the input neurons to improve biased estimations with L1-regularisation. Including shrinkage parameters and Huber’s M-estimation parameter, the hyperparameters are determined via grid search and cross-validation. To evaluate the proposed algorithm, simulations were conducted with both an artificial dataset and an industrial dataset from a practical gasoline treatment process. The results indicate that the proposed algorithm is superior in terms of predictive accuracy and robustness to the classic MLP and the regularised soft-sensing approaches LASSO-MLP and dLASSO-MLP.

Deployment of metabolic heat rate based soft sensor for estimation and control of specific growth rate in glycoengineered Pichia pastoris for human interferon alpha 2b production

Lack of appropriate process models, reliable online sensors, and process variability in bioprocess systems are poising challenges in real-time monitoring and control of critical process parameters (CPPs). This present investigation deals with the development of a non-invasive soft sensor by utilizing metabolic heat rate as input signal for online estimation of specific growth rate (μest) during the induction phase of glycoengineered Pichia pastoris for human interferon-alpha 2b (huIFNα2b) production. Feedforward strategy employing a predetermined exponential feeding of methanol during the induction phase was dealt at defined setpoint values (μSP). Standard PID controller with predetermined gain values regulated methanol feeding in accordance with the deviation from the μSP value. An adaptive PID (gain scheduling) significantly minimized the deviation of μ from its μSP value, reduced the amplitude of oscillation and achieved long-term controller stability. Robust control of methanol feeding by adaptive PID resulted in a 1.5 and 2.2-fold increase in productivity of huIFNα2b compared to standard PID and feedforward controls respectively. Moreover, adaptive PID control facilitated narrow range control of μ for longer durations (> 20 h) with a low average tracking error (< 6%) enumerating its scope of application in therapeutic protein production in near future.

Wireless Flexible Magnetic Tactile Sensor with Super-Resolution in Large-Areas.

Tactile recognition is among the basic survival skills of human beings, and advances in tactile sensor technology have been adopted in various fields, bringing benefits such as outstanding performance in manipulating objects and general human-robot interactions. However, promoting enhanced perception of the existing tactile sensors is limited by their sensor array arrangement and wire-connected design. Here we present a wireless flexible magnetic tactile sensor (FMTS) consisting of a multidirection magnetized flexible film (perception module) and a contactless Hall sensor (signal receiving module). The flexible magnetic film is composed of NdFeB microparticles and soft silicone elastomer microparticles, and it transfers the unambiguous transduction of external force position and magnitude into magnetic signals. Benefiting from the specific magnetization arrangement and clustering algorithm, only one Hall sensor is needed in FMTS to perceive the magnitude and position of the contact spot simultaneously with super-resolution (2.1 mm average error) on a large area (3600 mm2), and the effective working distance is also greatly extended (∼30 mm), allowing for the full softness and adaptability to diverse conditions. We anticipate that this design will promote the development of soft tactile sensors and their integration into human-robot interaction and humanoid robot perception.

Embedded Core-Shell 3D Printing (eCS3DP) with Low-Viscosity Polysiloxanes.

Flexible core-shell 3D structures are essential for the development of soft sensors and actuators. Despite recent advancements in 3D printing, the fabrication of flexible 3D objects with internal architectures (such as channels and void spaces) remains challenging with liquid precursors due to the difficulty to maintain the printed structures. The difficulty of such fabrication is prominent especially when low-viscosity polysiloxane resins are used. This study presents a unique approach to applying direct ink writing (DIW) 3D printing in a three-phase system to overcome this limitation. We performed core-shell 3D printing using a low-viscosity commercial polysiloxane resin (Ecoflex 10) as shell inks combined with a coaxially extruded core liquid (Pluronic F127) in Bingham plastic microparticulate gels (ethanol gel). In the process termed embedded core-shell 3D printing (eCS3DP), we highlighted the dependence of the rheological characteristics of the three fluids on the stability of the printed core-shell filament. With the core liquid with a sufficiently high concentration of Pluronic F127 (30 w/w%; σy = 158.5 Pa), the interfacial instability between the shell liquid and core liquid was suppressed; the removal of the core liquid permitted the fabrication of perfusable channels. We identified the printing conditions to ensure lateral attachments of printed core-shell filaments. Interestingly, judicious selection of the rheological properties and flow rates of three phases allowed the formation of droplets consisting of core liquids distributed along the printed filaments. eCS3DP offers a simple route to fabricate 3D structures of a soft elastomeric matrix with embedded channels and should serve as a useful tool for DIW-based fabrication of flexible wearable devices and soft robotic components.

Stream-based active learning with linear models

The proliferation of automated data collection schemes and the advances in sensorics are increasing the amount of data we are able to monitor in real-time. However, given the high annotation costs and the time required by quality inspections, data is often available in an unlabeled form. This is fostering the use of active learning for the development of soft sensors and predictive models. In production, instead of performing random inspections to obtain product information, labels are collected by evaluating the information content of the unlabeled data. Several query strategy frameworks for regression have been proposed in the literature but most of the focus has been dedicated to the static pool-based scenario. In this work, we propose a new strategy for the stream-based scenario, where instances are sequentially offered to the learner, which must instantaneously decide whether to perform the quality check to obtain the label or discard the instance. The approach is inspired by the optimal experimental design theory and the iterative aspect of the decision-making process is tackled by setting a threshold on the informativeness of the unlabeled data points. The proposed approach is evaluated using numerical simulations and the Tennessee Eastman Process simulator. The results confirm that selecting the examples suggested by the proposed algorithm allows for a faster reduction in the prediction error.

Detecting flooding state in extraction columns: Convolutional neural networks vs. a white‐box approach for image‐based soft sensor development

Among thermal separation methods, solvent extraction with countercurrent flow is one of the most energy-efficient processes. The separation efficiency is highest close to the flooding point, which characterizes the maximum possible liquid throughput. However, this point is the limiting parameter in the entire extraction process since the flow of at least one phase is stopped. The optical supervision of the equipment enables the efficient operation of the extraction column near the flooding state. This contribution presents an assessment of two approaches in detecting the undesirable flooding state through recorded images of a transparent extraction column. One approach is a feature-based space that is more interpretable in extracting features than a Support Vector Machine classifier. The second is a feature-extraction pipeline followed by a robust Convolutional Neural Network model. The analysis regarding the inference times shows that both models can be used as online classification tools for real-time applications.

Model stacking to improve prediction and variable importance robustness for soft sensor development

This paper presents an interpretable ensemble modelling method, in which the predictions of several individual base learners are combined together through Stacked generalisation, which makes use of a secondary layer model, or so called meta-learner, that is trained on the output cross-validation predictions of each base learner. To provide interpretability, the permutation variable importance (PVI) is computed on the ensemble, wherein variables are randomly shuffled and the reduction in predictive performance for the ensemble is calculated for each variable. This is a novel contribution, as no previous attempts have been made in the soft sensor literature to investigate the interpretability of ensemble models that use heterogeneous base learners. The Stacked ensemble model also avoids model selection, which is the process of choosing among many candidate models. Model selection is often based on cross-validation, which is not guaranteed to select the best model in terms of true generalisation performance on the test set. Instead, the proposed method combines multiple models instead of choosing a singular model, avoiding the need for model selection. The efficacy of the proposed methodology in terms of both variable importance and predictive performance is shown on a synthetic dataset, in which the variable importance is already known, and an industrial dataset of a refinery process provided by Dow. For the synthetic dataset, it is shown that the proposed method chooses the correct casual variables, whereas the in-built variable importance provided by the individual models, namely Partial least squares, Lasso, Random forests & XGBoost, can give increased importance to non-causal, randomly generated variables. For the industrial study, the combined ensemble is shown to outperform all individual base models in terms of predictive performance, whilst also providing a new perspective in terms of variable importance compared to previous studies.

Nonlinear Dynamic Soft Sensor Development with a Supervised Hybrid CNN-LSTM Network for Industrial Processes.

A soft sensor is a key component when a real-time measurement is unavailable for industrial processes. Recently, soft sensor models based on deep-learning techniques have been successfully applied to complex industrial processes with nonlinear and dynamic characteristics. However, the conventional deep-learning-based methods cannot guarantee that the quality-relevant features are included in the hidden states when the modeling samples are limited. To address this issue, a supervised hybrid network based on a dynamic convolutional neural network (CNN) and a long short-term memory (LSTM) network is designed by constructing multilayer dynamic CNN-LSTM with improved structures. In each time instant, data augmentation is implemented by dynamic expansion of the original samples. Moreover, multiple supervised hidden units are trained by adding quality variables as part of the layer input to acquire a better quality-related feature learning performance. The effectiveness of the proposed soft senor development is validated through two industrial applications, including a penicillin fermentation process and a debutanizer column.

Development of a machine learning-based soft sensor for an oil refinery’s distillation column

In this paper, a machine learning framework based on Kaizen Programming is proposed for building a soft-sensor using real historical data from an oil refinery. The soft-sensor estimates the composition of C4 hydrocarbons in the distillate stream of a splitter column. Kaizen Programming is a novel technique that has shown excellent results for symbolic regression problems without requiring a priori selection of the functional bases. The framework follows three different steps: pre-processing of the data, model building with the selected algorithm, and ensemble of different good-fitting models. The final ensemble model accurately predicts operation of the column at and between two steady states, outperforming the model based on Gaussian Process in both validation and soft-sensor degradation scenarios.

Soft Resistive Tactile Sensor Based on CNT-PDMS-Gel to Estimate Contact Force

Soft tactile sensors have been extensively studied due to their promising potential in estimating applied forces while exhibiting mechanical compliance. Such sensors have been fabricated from a variety of composite materials and various techniques to advance their applications. Currently, the development of soft tactile sensors based on carbon nanotubes (CNT)–polydimethylsiloxane (PDMS) gel has yet been investigated. This letter presents the design of a soft tactile sensor based on a CNT-–PDMS-gel composite that is sensitive to external forces. The sensor was composed of an elastomer, a rectangular conductive body made from a PDMS–CNT-gel composite, and a circuit board. The electrical resistance of the composite changed in response to the strain applied. Push and release tests were conducted, and the conductive body with 1% CNT was chosen due to high sensitivity, fast response, and stability in reading output. Experimental results, using a feed-forward neural network, showed that the sensor can estimate the applied normal force based on the changes of resistance.

Development of a wide-range soft sensor for predicting wastewater BOD5 using an eXtreme gradient boosting (XGBoost) machine

In wastewater monitoring, detecting extremely high pollutant concentrations is necessary to properly calibrate the treatment process. However, existing hardware sensors have a limited linear range which may fail to measure extremely high levels of pollutants; and likewise, the conventional “soft” model sensors are not suitable for the highly-skewed data distributions either. This study developed a new soft sensor by using eXtreme Gradient Boosting (XGBoost) machine learning to ‘measure’ the wastewater organics (in terms of 5-day biochemical oxygen demand (BOD5)). The soft sensor was tested on influent and effluent BOD5 of two different wastewater treatment plants to validate the results. The model results showed that XGBoost can detect these extreme values better than conventional soft sensors. This new soft sensor can function using a sparse input matrix via XGBoost's sparsity awareness algorithm - which can address the limitation of the conventional soft sensor with the fallibility of supporting hardware sensors even.

Mechanism Analysis and Self-Adaptive RBFNN Based Hybrid Soft Sensor Model in Energy Production Process: A Case Study.

Despite hard sensors can be easily used in various condition monitoring of energy production process, soft sensors are confined to some specific scenarios due to difficulty installation requirements and complex work conditions. However, industrial process may refer to complex control and operation, the extraction of relevant information from abundant sensors data may be challenging, and description of complicated process data patterns is also becoming a hot topic in soft-sensor development. In this paper, a hybrid soft sensor model based mechanism analysis and data-driven is proposed, and ventilation sensing of coal mill in a power plant is conducted as a case study. Firstly, mechanism model of ventilation is established via mass and energy conservation law, and object-relevant features are identified as the inputs of data-driven method. Secondly, radial basis function neural network (RBFNN) is used for soft sensor modeling, and genetic algorithm (GA) is adopted for quick and accurate determination of the RBFNN hyper-parameters, thus self-adaptive RBFNN (SA-RBFNN) is proposed to improve the soft sensor performance in energy production process. Finally, effectiveness of the proposed method is verified on a real-world power plant dataset, taking coal mill ventilation soft sensing as a case study.