Scholarly article on topic 'Assisting the Design of Sensor and Information Fusion Systems'

Assisting the Design of Sensor and Information Fusion Systems Academic research paper on "Computer and information sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Technology
OECD Field of science
Keywords
{"intelligent technical systems" / "cyber-physical systems" / "information fusion" / "fusion system design" / "intelligent sensors" / self-configuration / self-diagnosis / "intelligent networking" / "real-time communication"}

Abstract of research paper on Computer and information sciences, author of scientific article — Uwe Mönks, Henning Trsek, Lars Dürkop, Volker Geneiß, Volker Lohweg

Abstract The increased deployment of information technology for information processing, extensive networking, and system/environment monitoring using sensor and information fusion systems are essential characteristics of cyber-physical systems. They allow an autonomous recognition and evaluation of the system's status leading to autonomous reactions improving or maintaining the status to operate adaptively, robustly, anticipatory, and user-friendly. Assisting the operator in handling such complex systems is rather important and requires self-configuration, self-diagnosis, and self-optimization capabilities. In this paper, a new assisted design methodology for sensor and information fusion systems is proposed. It is based on an innovative system architecture consisting of the information fusion system itself, intelligent adaptable sensors, and the communication architecture of the “Intelligent Technical Systems OstWestfalenLippe” (it's OWL) Leading-Edge Cluster project “Intelligent Networking” providing an intelligent network for self-configuration and the required real-time data exchange.

Academic research paper on topic "Assisting the Design of Sensor and Information Fusion Systems"

Available online at www.sciencedirect.com

ScienceDirect

Procedia Technology 15 (2014) 35 - 45

2nd International Conference on System-Integrated Intelligence: Challenges for Product and

Production Engineering

Assisting the design of sensor and information fusion systems

Uwe Mönksa*, Henning Trseka, Lars Dürkopa, Volker Geneißb, Volker Lohwega

ainIT -Institute Industrial IT, Ostwestfalen Lippe University of Applied Sciences, Liebigstr. 87, 32657 Lemgo, Germany bFraunhofer ENAS, Warburger Straße 100, 33098 Paderborn, Germany

Abstract

The increased deployment of information technology for information processing, extensive networking, and system/environment monitoring using sensor and information fusion systems are essential characteristics of cyber-physical systems. They allow an autonomous recognition and evaluation of the system's status leading to autonomous reactions improving or maintaining the status to operate adaptively, robustly, anticipatory, and user-friendly. Assisting the operator in handling such complex systems is rather important and requires self-configuration, self-diagnosis, and self-optimization capabilities. In this paper, a new assisted design methodology for sensor and information fusion systems is proposed. It is based on an innovative system architecture consisting of the information fusion system itself, intelligent adaptable sensors, and the communication architecture of the "Intelligent Technical Systems OstWestfalenLippe" (it's OWL) Leading-Edge Cluster project "Intelligent Networking" providing an intelligent network for self-configuration and the required real-time data exchange.

© 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/3.0/).

Peer-review under responsibility of the Organizing Committee of SysInt 2014.

Keywords: intelligent technical systems; cyber-physical systems; information fusion; fusion system design; intelligent sensors; self-configuration; self-diagnosis; intelligent networking; real-time communication

1. Introduction

Mechatronic systems are currently subject to change towards intelligent technical systems. Recent years were marked by an increased application of information technology for information processing, extensive networking, and system as well as environment monitoring using sensor and information fusion (SEFU/IFU) systems. Thus, these systems become more intelligent in order to assist the operator in handling such complex systems.

Industrial printing processes, like a newspaper printing process depicted in Fig. 1, are typical applications in which the printing presses evolve from mechatronic to intelligent technical systems. Today's state-of-the-art printing systems are driven by hundreds of actuators in the application, along with a number of sensors in the same order of

* Corresponding author. Tel.: +49-5261-7025993; fax: +49-5261-702312. E-mail address: uwe.moenks@hs-owl.de

2212-0173 © 2014 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/3.0/).

Peer-review under responsibility of the Organizing Committee of SysInt 2014. doi:10.1016/j.protcy.2014.09.032

magnitude to acquire different types of data. Hence, thousands of parameters are to be supervised for the determination of a printing press' state, but each set of supervised parameters changes from one printing press to another, because every press is highly customized. In every case, the sensors must be chosen and parameterized properly (e.g., regarding the appropriate measurement range). Furthermore, novel sensory concepts which turn actuators into sensors get introduced nowadays, e.g., the Motor as Sensor (MaS) concept [2].

Fig. 1. KBA Cortina newspaper offset printing press (with kind permission of KBA - Koenig & Bauer AG, Wurzburg [1]).

Currently, it is due to the sheer number of data sources a complex, time consuming, and error-prone task to provide such systems manually with self-diagnosis abilities in form of a SEFU/IFU system: The system developer has to select, parameterize, operate, and validate problem-adjusted sensors and evaluation algorithms (for signal acquisition and preprocessing) as well as possible additional information sources. Since the number of possible solutions is most frequently too vast, the developer must be very experienced and will probably create a system based on already existing solutions which in only very few cases represents the optimal solution. Furthermore, setting up a communication network with real-time capabilities is also required comprising several manual steps. The project "Intelligent Networking" [4] deals with self-configuration of both the communication and the fusion system enabling self-diagnosis for self-optimization, hence autonomous recognition and evaluation of the technical system's status leading to autonomous reactions improving or maintaining the status.

This paper is structured as follows. Related work and existing research gaps are discussed in section 2. Based on these findings and the identified challenges of the introduction, an innovative information fusion system architecture and a corresponding design assistance are proposed in section 3. In section 4 the architecture and the design flow are preliminary evaluated in the context of a real setup. The paper is concluded in section 5 followed by a brief outlook towards future work.

2. Related work

The main challenges described in section 1-or similar challenges-have also been tackled by others. Work leading towards first partial solutions for intelligent technical systems can be categorized in the three classes, (i) sensor and information fusion, (ii) intelligent sensors, and (iii) self-configuration of networks, which are discussed accordingly.

Sensor and information fusion systems appear in various kinds. They all have in common, that the information originating from a number of homogeneous or heterogeneous sources are combined in order to reduce the data's amount and dimensionality, and obtain information of higher quality. Comprehensive overviews over information sources, general fusion system models, and theoretical backgrounds of the fusion algorithms provides [5].

For modeling the sensor information and carrying out the fusion, one of the evidence theories (probability, possibility, belief theory) as well as concepts derived from them are applied most likely [6], [7], [8]. We identified a hybrid approach consisting of a possibility theory-based information model and a belief theory-based fusion as beneficial for machine condition monitoring applications of various kinds [9], [10]. This fusion system, referred to as multi-layer attribute-based conflict reducing observation system MACRO, is capable of resembling the physical

structure of the monitored application in the fusion system, and determines and reduces conflicts between information sources.

As briefly mentioned in section 1, the design of a fusion system is mainly a time-consuming manual task carried out by an expert. To tackle this issue, Iswandy and König present a framework for the automated design of multisensor fusion systems in [11]. Nevertheless, the proposed framework follows a sequential optimization procedure which aims at replacing the human expert. But as Hall makes clear, a human expert understanding the monitored system or process to evaluate the decisions made by the fusion system is necessary in every case [12]. Therefore a system assisting the design and leaving the ultimate decision to the user is suited better, instead. Furthermore, Iswandy and König's framework is not publicly available.

Today, intelligent sensors are almost not available and automation applications use almost solely elementary sensors without an own signal preprocessing. The parameter range is physically defined and sent as an analog signal with linearity error and offset to the evaluation unit [13]. For further digital processing the full available measuring range of the elementary sensor is digitized with a defined resolution. An adaptation of the resolution with respect to the required measuring range of the application is impossible [14]. Nowadays, additional signal processing units are often combined with the real sensor to hybrid sensor modules [15]. For an improvement of the sensor sensitivity these systems use band-pass filtering and signal averaging of multiple measurements [16]. More complex signal analysis is usually performed in a centralized signal processing unit delivering an amplified, offset-corrected and linearized analog output signal [17]. Only in rare cases an analog/digital converter (ADC) is also part of the module [18]. Nevertheless, the complete physical measurement range of the sensor is typically also digitized with a fixed resolution in these modules. Decreasing the analog input range to improve the digital resolution is impossible. Monolithic integrated sensor systems increasingly contain sensor, amplifier, ADC, and signal processing logic including filter components on a single chip [19], but they are not equipped with suitable interfaces. Furthermore, every sensor is designed to capture only one environmental parameter (e.g., temperature, magnetic field, humidity, gas concentration, pressure) and thus a software-based selection of the measurement parameter is not possible. To sum up, no intelligent adaptive sensor system exists, that is able to grasp a certain environmental or state variable and automatically adapt the measurement range to its current conditions.

Intelligent networks and self-configuration were already the objective of some research projects in order to provide plug-and-play (PnP) functionalities to arbitrary nodes and modules of a system. In SOCRADES a service-oriented architecture (SOA) on device level was developed [20]. The project proposed a framework for flexible service orchestration based on petri nets. However, every reconfiguration step results in manual effort. In order to achieve the objective of PnP, the system must react autonomously to changes. The EU project Internet of Things at Work (IoT@Work) [21] specified a secure plug-and-work communication infrastructure taking industrial automation requirements into account and supporting autonomous network configuration and optimization. However, the project focused only on the lower layers of the communication system. Finally, the project AutoPnP [22] dealt with automatically adding and configuring new components in automation systems, based on artificial intelligence concepts and algorithms. However, the project focused on the control level including manufacturing execution systems (MES) with decreased temporal requirements. Besides, several individual research works were directed towards self-configuration covering various relevant topics in this area. A semantic self-description is in most cases the basis to implement self-configuration of intelligent technical systems. In [23] the communication between different manufacturing modules is analyzed and, based on the findings, a lightweight taxonomy to solve the problem of signal identification is introduced. The taxonomy is evaluated with a real implementation based on industrial standards such as OPC UA [24]. Another important aspect of self-configuration is the automatic configuration of real-time communication networks which usually require numerous manual configuration steps. A promising approach is presented in [25], [31] based on Profinet and in [26] based on Powerlink. Both works divide the configuration in different steps and have in common that they need an ad hoc channel for the configuration. However, none of the existing research works discussed in this section is able to provide deterministic real-time communication with low latency and jitter and at the same time supporting self-configuration. Therefore, a new communication architecture with self-configuration capabilities has been developed and integrated into intelligent sensors as well as into the fusion system to facilitate a complexity reduction during setting up such systems.

3. Sensor and information fusion system design

In order to address the existing problems mentioned in section 1, propositions for a new design methodology for sensor and information fusion systems are presented in this work. The methodology is based on an innovative system architecture. Besides the information fusion system itself, it consists of intelligent adaptable sensors, and an intelligent network providing the required real-time data exchange. The methodology addresses the challenges for implementing a development framework for resource-limited intelligent technical systems which instructs and supports the developer to generate SEFU/IFU systems aiming at reducing the fusion system's complexity and finding a problem-adjusted solution. For this purpose the methodology uses information about the prevailing problem to generate an optimal overall solution under the given constraints based on a multitude of possible partial solutions, and to propose additional presumably less optimal alternative solutions. Thus, the developer is supported in generating complex SEFU/IFU systems, but finally decides which proposal will be applied.

3.1. Architecture

The information fusion system architecture (cf. Fig. 2) consists of several building blocks considered as intelligent nodes, such as intelligent sensors, a central signal processor, and an intelligent network for their interconnection.

Fig. 2. Sensor and information fusion system architecture and components.

The intelligent sensors provide the interface to the actual environment, i.e., the machine to be monitored. An intelligent sensor module could be equipped with various types of environmental sensors, e.g., visual sensors, acoustic sensors, physical sensors (temperature, pressure, etc.), each measuring different physical parameters. The required parameters can be selected and the sensor is able to adjust its parameters and range based on the environment. Every intelligent sensor node, hence being an implementation in the sense of cyber-physical systems [27], is equipped with fusion capabilities to (i) reduce data amount and dimensionality, (ii) extract information from the acquired data, and (iii) enable adaptation and self-diagnosis capabilities of the intelligent nodes regarding conflicting sensor values.

Moreover, every component of the system is connected via an intelligent network. The network supports mainly three important characteristics, (i) semantic descriptions, (ii) decoupling of the application and the communication system, and (iii) real-time communication including its autoconfiguration. The semantic description of an intelligent sensor belongs to the application layer. The middleware decouples the application layer from the connectivity and offers different services to the application and the connectivity. Finally, the connectivity layer provides deterministic real-time communication allowing all components to exchange their process data within a pre-defined amount of time and an ad hoc channel which is used for configuration purposes.

The information acquired by each intelligent sensor are transferred via the intelligent network to the central signal processor. This device is, like the intelligent sensors, equipped with communication and fusion capabilities, but likely being more powerful. Here, the global fusion of the intelligent sensors' information is carried out for the state determination of the system or process being monitored. It is operated by an integrated or connected human machine interface. All components are described in detail in the consecutive subsections.

3.2. Information fusion

The architecture depicted in Fig. 2 presents an application point of view on the fusion system. It is independent from any fusion technique. We propose the multi-layer attribute-based conflict-reducing observation approach MACRO (cf. Fig. 3) for machine condition monitoring applications as described in section 1.

Fig. 3. Multi-layer attribute-based conflict-reducing observation system MACRO.

For the determination of the global state of a complete system or process, signals of itself as well as its environment are acquired by sensors (signal sources). Features are extracted from the signals in the following signal conditioning step which may also include signal preprocessing procedures. Ensembles of conditioned signals are then grouped to so-called attributes representing certain properties or physical parts of the observed system or process. The attributes are application-dependent and defined during the design process. An example of such a property called attribute in this context is the force a rotating cylinder in a printing press is exerting to another rotating cylinder. The force can be measured directly by a pressure sensor, but also indirectly by evaluating features of solid borne. Redundancies occurring by combining at least two information sources to one attribute are used beneficially for both (i) intercepting sensor faults and (ii) cross-checking the consistency of sensor values. The latter is carried out implicitly by the psychologically inspired fuzzified balanced two-layer conflict solving (mBalTLCS) fusion approach [9]: mBalTLCS creates one output signal per attribute from its input signals and assigns the attribute an importance measure which is the negated conflict between the sensors' individual opinions. Conflict occurs whenever information does not bear evidence for only one opinion/proposition, but also for another. Subsequently, the fused attributes' opinions (mBalTLCS output signals) are aggregated on system level using the Implicative Importance Weighted Ordered Weighted Averaging (IIWOWA) operator [28] to reason about the complete system or process under supervision. It weights each attribute according to its previously determined importance, such that attributes with a low conflict have a higher impact on the aggregation result. Detailed information regarding MACRO can be found in [9].

The mapping onto the general architecture depicted in Fig. 2 is not bound tight. Though, the most bottom and the most top layers are fixed: the applied N sensors belong to n < N intelligent sensors while the aggregation on system layer is carried out on the central signal processor which determines the current global state. Everything in between is either implemented in the central signal processor, or distributed over the network of intelligent sensors. In the first case, each intelligent sensor solely acquires the signals in the desired sufficient quality from the attached sensors (cf. section 3.3) and transmits them to the central signal processor. Here, all following steps are implemen-

ted and carried out, containing further signal conditioning steps (like preprocessing and feature extraction) and ßBalTLCS fusion on attribute layer. This procedure results in the highest possible amount of data communicated over the network on the one hand, on the other hand the computational efforts concentrate on the central signal processor. Hence, no scalable solutions are possible as the network's capacity is exhausted already by some ten sensor nodes (cf. [2]). Instead, we propose to prefer handling all processing steps including attribute layer fusion by the intelligent sensors in a distributed manner. This results in a minimum amount of data being communicated over the network, as only the attributes' results must be transmitted to the central signal processor.

MACRO already showed its capabilities in various machine condition monitoring applications like during a plastic pipe extrusion process [9] or ATM monitoring [10]. Hence, MACRO as well as other SEFU/IFU systems are involved in varying applications with varying requirements and constraints. These include sensor configurations, choices for, e.g., signal preprocessing algorithms or features, attribute definitions, and parameter settings. Thus, although the MACRO system is fixed in its structure, it still has many degrees of freedom. In the following we describe the possibilities assisting a SEFU/IFU system designer during the creation and operation phase.

As already mentioned, the design of multisensor and information fusion systems is a time-consuming and complex manual task. It involves expertise in the fields of sensors, signal processing, computational intelligence, and software design, besides the inevitable knowledge about the system or process to be monitored. Usually, the expert relies on his or her experience and thus a SEFU/IFU system to the best knowledge of the designer is created which might not be the optimal solution. Since the designer is limited in resources, not all possible combinations can be evaluated and considered in the SEFU/IFU system to be created. Additionally, manual designs are prone to error due to, e.g., human failure and missing testing time. A possibility to overcome the design issues is the introduction of a completely automated design process as proposed by Iswandy and König [11]. Their framework relies on a knowledge base containing a pool of sensors and methods together with their initial parameters which are sequentially optimized during the design process for creating a fusion system. The framework makes use of genetic algorithm and particle swarm optimization approaches for parameter optimizations and algorithm choices [29], [30]. We consider such a design tool involving local optimization algorithms heading into the right direction, but due to the experiences made in various applications, we believe relying only on the output of an automatic design framework is the wrong approach. Especially when it comes to robustness, a human expert is inevitably necessary for the final evaluation of a proposed SEFU/IFU system. Additionally, in the end the system must be transparent and understandable for a human expert such that erroneous situations can be resolved. Thus, our requirements to a design methodology are the following, from which parts have already been solved:

• Each application demands appropriately chosen sensors matching the application's requirements with respect to the measured quantity, the measurement range and resolution. Instead of testing different sensors and their parameters, we suggest the use of a general intelligent sensor as described in section 3.3. It is able to adapt to varying conditions, guaranteeing to operate in the best possible configuration. The intelligent sensor is on top able to describe itself in an appropriate language and form, e.g. in terms of its location and measured quantities.

• A SEFU/IFU design methodology shall serve as an assistance to the system designer. The human expert must remain the last instance in the decision upon the fusion system to be chosen. An automated framework, like the integrated one introduced by Iswandy and König [11], may at most suggest possible solutions from which the system designer picks the most appropriate one.

• As is commonly known, nothing like a gold standard for SEFU/IFU systems exists and each fusion system is created application-dependently. Nevertheless, tasks may be comparable from one application to another such that partial solutions may be applied again, rather than starting with a completely new fusion system design. Consequently, a design system must be able to store problem formulations and corresponding solutions in an appropriate way such that it can recognize these similarities.

• The signals grouped to attributes of the MACRO approach bear enough descriptive information to generate, update, and destroy the attributes automatically. These mechanisms are already available and working such that the designer does not need to spend effort on this task. We demonstrated the mechanisms in an ATM monitoring application, where information available in a proprietary form were processed [10]. The SEFU/IFU design methodology shall instead extend the available autoconfiguration mechanisms such that they are able to process self-description data (available in an appropriate language and form) originating from the intelligent sensors and the intelligent network (cf. section 3.4, necessary for the automatic determination of the physical topology).

All considerations regarding the SEFU/IFU design must be carried out under the constraining resource limitations evident in cyber-physical system implementations. In the following sections we describe necessary results in an intermediate state which are needed for the success of SEFU/IFU system design assistance.

3.3. Intelligent sensors

Since no intelligent adaptive sensor system exists (cf. section 2), one has been designed that on the one hand grasps a certain environmental or state variable. On the other hand, it can automatically adapt the measurement range to the current conditions and therefore collect high-resolution measurement values. The data are preprocessed and passed to a signal processing unit, together with information about its location and measurement range. Besides, the parallel capture of various system parameters by the intelligent sensor is also meaningful to gather system errors or faulty sensors in situ by signal fusion and to possibly compensate them accordingly. To communicate with other devices in the fusion system (such as the central signal processor) the intelligent sensor also uses the communication architecture as described in section 3.4. As basic architecture of the intelligent sensor module an electronic circuit has been developed and is shown in Fig. 4 (b). It processes the sensor signal by an instrumental amplifier, and digitizes the measured signals by a 10 bits analog digital converter (ADC). The I2C interface of the microprocessor allows a flexible variation of the reference voltage of the internal ADC and thus the resolution spectrum of the converter is extended. In the first development stage of the sensor interface, resistive sensors and thermocouples are considered and evaluated to develop a suitable calibration procedure.

A system prototype is based on the Arduino Uno platform. The desired dynamic measurement range extension for the adaptation of the digital resolution is achieved by an extended conversion of the ADC. In normal operation mode, the onboard ADC of the Arduino (microcontroller AVR Mega328P) can resolve the analog measurement voltage with 10 bits (1024 steps). In this case, the analog reference voltage of the ADC supplies the full-scale voltage level. The maximum resolution of the ADC is achieved with the lowest possible reference voltage, which is in this case minimally 1.1V. By a software-controlled adaptation of this reference voltage, the intelligent sensor can be adjusted to various sensors like thermocouples, pressure, and brightness sensors. Therefore, an external digital analog converter (DAC) is connected to the analog reference voltage port of the Arduino Uno, controlled via its I2C interface. The 4.1 V fixed reference voltage source MCP1541T is used as high-precision reference voltage. This design permits a maximum resolution of 1.1 V/1024, hence approximately 1 mV. This setup allows the investigation of the intelligent software-based adaption of the sensor hardware to varying sensor signal inputs. Further development for a wider adaption range of the analog input signals (auto range functionality) will be developed in future.

3.4. Intelligent communication network

An intelligent communication network is one of the enablers to ease the design of sensor and information fusion systems. Hence, the specified reference architecture of the project "Intelligent Networking" [4], [25] is integrated into all system components, i.e., the signal processor and the intelligent sensors. It follows a layered approach and is shown in Fig. 4 (a). The architecture has the main objective to provide services for a self-configuration of the realtime communication network and to decouple the application from the communication network. Therefore, three architectural layers have been defined: application, middleware, and connectivity.

Fig. 4. Communication architecture for intelligent networking of information fusion systems (a) and intelligent sensor node (b).

The application layer usually provides the required functionality for the system. In this work, it may consist of sensor objects and fusion blocks. The sensor objects can be considered as a software abstraction of the real physical application process, i.e., application process objects. Every sensor object is composed of the sensor signal data itself and a semantic description of the object and the data. A basic requirement for implementing self-configuration is a formalism to describe the sensor objects and other process objects. On the one hand, this encompasses a description language which can be interpreted by the middleware layer. On the other hand, the data and messages of the application layer objects and their functions must be semantically described. For instance, the semantic description of a simple temperature sensor could include the measured unit, its range, and other relevant information for the application such as its spatial position. The fusion component is part of the multi-layer information fusion system MACRO and has basically two tasks. First, it collects data from sensors and other information sources, which can be either directly connected to the node or logically belong this component. Second, the gathered signals are grouped on the next higher layer to an attribute, its values are fused, and subsequently passed to the centralized signal processor. A detailed discussion about information fusion and relevant algorithms is provided in subsection 3.2.

The middleware layer is responsible for decoupling the application layer and the underlying connectivity layer as well as selecting a suitable communication technology. It offers a standardized communication interface to the application, independent of the underlying real-time communication systems. The middleware consists of several services being responsible for application data exchange and self-configuration of the nodes. For the data exchange the middleware has to select a suitable communication technology which meets the application requirements. The requirements are described within each application object and provided via a standardized interface. The self-configuration core component is a discovery service which allows to discover existing other nodes and recognize their semantics. For both functionalities it requires an ad hoc communication channel as described below. Moreover, the information model defines the semantic description of application layer objects. An example bootstrapping would consist of three steps: (i) discovery of other nodes, (ii) configuration of the real-time communication systems, and (iii) logical interconnection of sensor objects to their relevant peers. After these steps are completed, the routing block is responsible for selecting valid paths to exchange information between the nodes, especially if several realtime communication systems are used. Finally, the security block enforces application requirements regarding security, including measures ensuring data integrity and encryption to protect critical data against attackers and eavesdroppers.

The connectivity layer is capable of providing different real-time communication channels depending on the requirements of the application. They are used to exchange process data of the application with deterministic temporal boundaries. Besides real-time traffic, it is also possible to exchange best effort traffic which is used for any non-critical data such as monitoring data or network management. The connectivity layer also provides an ad hoc communication channel. As soon as the intelligent node is physically connected to the network, the ad hoc channel is automatically established and can be used to exchange configuration data during the bootstrapping phase of the intelligent node. The ad hoc channel is essential for the self-configuration of the communication, because the performed self-configuration procedures mainly rely on it [31]. Moreover, depending on the requirements of the

application, the connectivity layer is able to offer various communication technologies ranging from deterministic wired hard real-time communication systems with cycle times < 1 ms to wireless technologies providing only soft real-time guarantees.

4. Preliminary evaluation

Finally, a preliminary qualitative evaluation was conducted in the context of a roller demonstrator setup being part of the Lemgoer Modellfabrik (www.smartfactory-owl.de). The demonstrator represents a small part of a printing machine as introduced in section 1. Situations representing erroneous machine operating points, faulty sensors, but also permitted varying production conditions can easily be simulated. The demonstrator will be used in our future work for more extensive case studies. It allows testing and comparing sensor fusion methods for machine analysis regarding their robustness and process real-time capability in a real scenario. Various sensors have been attached to the rotating print cylinder allowing the acquisition of analog and digital signals for temperature, acoustic emission, force, and others by a measurement data acquisition board. The setup can be easily extended by various intelligent sensors and the intelligent networking architecture. All signals are processed on a central device computing the machine's condition. The preliminary evaluation was designed to provide first qualitative insights into expected performance gains with respect to the current manual design procedure and in terms of relevant metrics. Those metrics, deciding upon the benefit of a solution and determining its acceptance in the market, are summarized in Table 1.

Fig. 5. Roller demonstrator for information fusion system evaluations (with kind permission of inlT - Institute Industrial IT, Lemgo).

Each metric is evaluated by o, +, or ++, where o represents least and ++ represents most benefit. The evaluation is based on a MACRO implementation in the roller demonstrator which has been set up manually using ordinary hard-wired sensor units. From the experiences made during the design, implementation, and commissioning processes (including pitfalls like inappropriate sensor parameterizations, cf. Table 1) the results for the assisted design approach incorporating intelligent sensors are extrapolated. Hence, the assisted design process has not been carried out, but effects from, e.g., intelligent sensor nodes with a well-defined communication interface are clearly imaginable. We assume that an assisted fusion system design process will have beneficial effects on the majority of the relevant metrics. Intelligent sensor nodes, for example, need no parameterization and can be applied in a plug-and-produce manner. This will lead to significantly less time consumption during commissioning and adds no engineering efforts regarding reconfiguration. By their ability to adjust autonomously to unforeseen situations and varying conditions, the robustness of monitoring systems towards such changing environments increases. Additionally, an assisted design process inherently includes a controlling instance which decreases the probability of errors to occur. The assistance system is capable to evaluate more possible solutions than an engineer can manually do, so the resulting solution is likely to be more problem-optimal compared to manual designs. Only the understandability of the solution will be much better for a fusion system expert with manual design as the solution is created stepwise by the expert.

Table 1. Preliminary evaluation-first qualitative results.

Metric Manual design process Assisted design with intelligent sensors

Time for commissioning O ++

Reconfiguration effort O ++

Error probability + ++

Robustness + ++

Optimality of the solution + ++

Understandability of the solution ++ +

5. Conclusion and outlook

In this contribution the challenges arising in the design process of sensor and information fusion systems for condition monitoring applications of cyber-physical system were presented. We described measures to be taken to counter the described challenges which led to a SEFU/IFU system architecture comprising intelligent sensors, an intelligent communication network, and the MACRO fusion system, with self-configuration mechanisms making use of self-descriptive information. But still, there are several open issues to be addressed in setting up the described methodology. But as the effort regarding these is only spent once, the following implementations are easier to be designed and set up. The break-even will be reached quite fast, especially in cases where a large number of SEFU/IFU systems need to be designed. Other use-cases are customer-specific productions like ATM production where a large number of model variants are produced. Currently, each ATM monitoring system must be parameterized manually depending on the current variant, whereas in case of intelligent sensor units, the fusion system orchestrates itself depending on the actual configuration.

Acknowledgements

This work was partly funded by the German Federal Ministry of Education and Research (BMBF) within the Leading-Edge Cluster "Intelligent Technical Systems OstWestfalenLippe" (it's OWL).

References

[1] Koenig & Bauer AG. Press images; 2014 [cited 2014 Feb 20]. Available from: http://www.kba.com/en/downloads-glossary/downloads/press-images/.

[2] Mönks U, Bator M, Dicks A, Lohweg V. Informationsfusion mit verteilter elektromotorischer Sensorik im Maschinen- und Anlagenbau (Information Fusion by Distributed Electric Motor Sensors in Machinery and Plant Engineering). In: Entwurf mechatronischer Systeme: 9. Paderborner Workshop. Paderborn: Heinz Nixdorf Institut; 2013. p. 305-16 (HNI-Verlagsschriftenreihe; vol. 310).

[3] BITKOM, editor. Leitfaden Big Data im Praxiseinsatz - Szenarien, Beispiele, Effekte (Guideline Big Data Applied in Praxis - Scenarios, Examples, Effects). Berlin; 2012 [cited 2014 Feb 20]. Available from: http://www.bitkom.org/de/publikationen/38337_73446.aspx.

[4] itsowl-IV: Intelligente Vernetzung (Intelligent Networking) [cited 2014 Feb 20]. Available from: URL:http://www.its-owl.de/technologiecluster/projekte/QP_Intelligente_Vernetzung.php.

[5] Hall DL, Llinas J, editors. Handbook of multisensor data fusion. Boca Raton, FL: CRC Press; 2001.

[6] Bishop CM. Pattern recognition and machine learning. Corr. at 8. print. New York, NY: Springer; 2009. (Information science and statistics).

[7] Zadeh LA. Fuzzy sets as a basis for a theory of possibility. Fuzzy sets and systems 1978; 1:3-28.

[8] Shafer G. A Mathematical Theory of Evidence. Princeton, NJ: Princeton University Press; 1976.

[9] Mönks U, Voth K, Lohweg V. An Extended Perspective on Evidential Aggregation Rules in Machine Condition Monitoring. In: 3rd International Workshop on Cognitive Information Processing (CIP 2012): IEEE; 2012. p. 1-6.

[10] Mönks U, Priesterjahn S, Lohweg V. Automated Fusion Attribute Generation for Condition Monitoring. In: Proceedings 23. Workshop Computational Intelligence. Karlsruhe, Baden: KIT Scientific Publishing; 2013. p. 339-53.

[11] Iswandy K, König A. Methodology, Algorithms, and Emerging Tool for Automated Design of Intelligent Integrated Multi-Sensor Systems. Algorithms 2009; 2(4):1368-409.

[12] Hall DL, Steinberg A. Dirty Secrets in Multisensor Data Fusion. Defense Technical Information Center Online 2001 [cited 2014 Feb 20]:1-15. Available from: http://www.dtic.mil/docs/citations/ADA392879.

[13] Mukopadhyay S C, Huang Y-M. Advancements in Modeling, Design Issues, Fabrication and Practical Applications. Sensors. Springer; 2008.

[14] Regtien PPL. Sensors for Mechatronics. London, Waltham MA USA: Elsevier; First edition 2012.

[15] Hilleringmann U. Mikrosystemtechnik, Wiesbaden: Vieweg-Teubner, GWV Fachverlage GmbH; 2006.

[16] Wiegand C, Hangmann C, Hedayat C, Hilleringmann U. A Resonance PLL-based Tracking System for Capacitive Sensors - MEMS/NEMS, SSI 2010: European Conference and Exhibition on Integration Issues of Miniaturized Systems, March 23-24 2010, Como, Italy.

[17] Boginski VL, Commander C, Pardalos PM. Sensors: Theory, Algorithms, and Applications. Berlin, Heidelberg: Springer; 2012.

[18] Islam SK, Haider MR. Sensors and low power signal processing. Berlin, Heidelberg: Springer; 2009.

[19] Suarez DF. Smart sensors and sensing technology. Nova Science Pub. Inc.; 2011.

[20] Colombo A W, Karnouskos S, Mendes J M. Factory of the Future: A Service-oriented System of Modular, Dynamic Reconfigurable and Collaborative Systems. In: Artificial Intelligence Techniques for Networked Manuf Enterprises Management, pp. 459-481. Springer, 2010.

[21] Houyou A M, Huth H P, Kloukinas C, Trsek H, Rotondi D. Agile Manufacturing: General Challenges and an IoT@Work Perspective. In: 17th IEEE Int Conference on Emerging Technologies and Factory Automation (ETFA 2012); Krakow, Poland, 2012.

[22] AutoPNP consortium. AutoPNP - Plug and Play for Automation Systems 2014 [cited 2014 Feb 20] Available from: http://www.autopnp.com/.

[23] Otto J, Böttcher B, Niggemann O. Plug-and-Produce: Semantic Module Profile. In: Dagstuhl-Workshop MBEES: Modellbasierte Entwicklung eingebetteter Systeme IV; 2013.

[24] Damm M, Leitner S, Mahnke W. OPC Unified Architecture. Springer-Verlag Berlin Heidelberg, 2009.

[25] Dürkop L, Trsek H, Jasperneite J, Wisniewski L. Towards Autoconfiguration of Industrial Automation Systems: A Case Study Using PROFINET IO. In: 17th IEEE Int Conference on Emerging Technologies and Factory Automation (ETFA 2012); Krakow, Poland, 2012.

[26] Reinhart G, Krug S, Huttner S, Mari Z, Riedelbauch F, Schlogel M. Automatic configuration (plug & produce) of industrial ethernet networks. In 9th IEEE/IAS International Conference on Industry Applications (INDUSCON), pages 1 - 6, 2008.

[27] acatech - Deutsche Akademie der Technikwissenschaften e.V., editor. Cyber-physical systems: Driving force for innovation in mobility, health, energy and production. Berlin: Springer; 2011. (acatech POSITION PAPER).

[28] Larsen HL. Importance weighted OWA aggregation of multicriteria queries. In: 18th International Conference of the North American Fuzzy Information Processing Society (NAFIPS 1999); 1999. p. 740-4.

[29] Corcoran P, Anglesea J, Elshaw M. The application of genetic algorithms to sensor parameter selection for multisensor array configuration. Sensors and Actuators A: Physical 1999; 76(1-3):57-66.

[30] Escalante HJ, Montes M, Sucar LE. Particle Swarm Model Selection. J. Mach. Learn. Res. 2009; 10:405-40.

[31] Dürkop L, Imtiaz J, Trsek H, Wisniewski L, Jasperneite J. Using OPC-UA for the Autoconfiguration of Real-time Ethernet Systems. In: 11th International IEEE Conference on Industrial Informatics (INDIN 2013); Bochum, Germany, 2013.