Bond Graphs as Mechatronic Approach for Supervision Design of Multisource Renewable Energy System

This paper deals with the bond graph methodology as a decision-making tool and multiphysic approach for supervision supervision system design. The developed algorithms have been applied to a mechatronic represented by Hybrid Renewable Energy System (HRES) which consists of Solar Photovoltaic Panels (PV) and wind turbines coupled to an electrolyser for hydrogen production used an energy vector for a fuel cell.


Introduction
Supervision is a set of tools and methods allowing the control of industrial processes in normal working as well as in presence of failures. At FDI (Fault Detection and Isolation) and FTC (Fault Tolerant Control) levels of the supervision and in the case of incidental or accidental situations, two kinds of information have to be displayed to operators. The first one concerns the detection and isolation of the fault, and the second deals with the way of continuing to control the process (reconfigurability). In order to help operators in those tasks, the integration of decision-making tools in the supervision system is needed. On one hand, supervision is associated with alarm processing to insure the safety based on dynamic model for fault indicators generation for online FDI. On the other hand the from human point of view, supervision system is seen by the operators in terms of functions and services (or missions) to be achieved. The set of services is organised into coherent subsets, called a User Operating Mode (USOM). At each time, and according to the technical specifications, a system carries on a coherent subset of missions. Each of these subsets is called operating mode. The supervision platform can be seen thus as a finite automaton G(A,) where A is a nodes representing a set of services and  is the Boolean transition from one node to another controlled by the FDI system: a hardware failure (provided by FDI layer) implies the unavailability of some services and could put back the pursuit of some missions in question [1]. This paper deals with the BG methodology [2] as a decision-making tool for supervision plateform design. The BG as a unified, multi disciplinary and understandable approach is well adapted for those tasks. Indeed, in the normal situation the BG model shows clearly to the operator the physical phenomenon and the power flowing in the process [3]. Furthermore, in incidental or accidental situations, the causal properties of the bond graph allow the operator to resolve problems related with the safety of the supervised process. Compared with classical approaches [4] such as the hybrid automata, the Operating Mode Management (OMM) is much effective due to the separation between the continuous dynamics, represented by the BG continuous state, and the discrete state governed by a classical automaton. The paper is organised as follows: The first part of the paper presentsBG as integrated design for not only dynamic modelling but also for supervision systems in terms of FDI. The second part concerns the real application of developed methodology to a complex HRES applied to a Hybrid Renewable Energy Systems (HRES) which consists of Solar Photovoltaic Panels (PV) and wind turbines coupled to an electrolyser for hydrogen production used an energy vector for a fuel cell. and the third section concludes the paper.

Bond graph theory for modelling and diagnosis
Complex mechatronic systems are considered as a combination of different energy components which are exchanging several powers. A BG which is a graph, G(S;A) is a unified graphical language for multi-physic domains. The nodes S represent physical components, subsystems, and other basic elements called junctions. While the edges A, called power bonds represent the power exchanged between nodes. This power is labelled by two conjugated power variables, named effort (e), and flow (f ). The set of components named BG element is: The R-element represents a passive energy dissipation phenomenon, while C, and I model the passive energy storage elements. (Se), and (Sf) are the sources of effort and flow, respectively. Sensors are represented by flow (Df), and effort (De) detectors. Finally, J, is used to connect the elements having the same effort (1-junction), or flow (0-junction). The conservative energy laws are obtained from the latter. TF, and GY are used to represent transformers and gyrators, respectively. From the BG model state space equations can be automatically generated, dedicated softwares are used for it. As shown in Figure 1(a), the power exchanged between two systems A and B indicated by a bond is the product of two variables -a potential variable (e.g. pressure, electrical potential, temperature, chemical potential, force, etc.) called effort (e) and a current variable (e.g. volume flow, current, entropy flow, velocity, molar flow, etc.) referred to as flow (f). One important structural property of the bond graph is its causality concept (algorithmic level of modelling). In the BG, it is denoted by the cross-stroke on the right indicating that the effort acts to the right, the side of the cross-stroke while the flow is in the reverse direction. As example in Figure 1(a), assigned causality means that system A imposes efforts on B. In the corresponding block diagram given by the Figure 1(b), the direction of action is indicated by an arrow on each connection as illustrated. Independently of the causality, the direction of the positive power is indicated by the half-arrow on the bond. Bond Graph modelling technique has been extensively used owing to the behavioural, structural and causal properties, that provide a systematic approach towards development of supervision and fault detection and Isolation (FDI) of highly non-linear and complex thermo-chemical systems. In BG framework, the model based FDI is mainly based upon ARRs as developed in [2].

Description of the process
The proposed approach is applied to model a small size experimental HRES presented in Fig. 9b. The system main objective is to produce hydrogen. The system is composed of two sources Photovoltaic Panel (PV) and Wind Turbine (WT) connected through a common DC bus to a battery, an Electrolyzer (EL) and a Proton Exchange Membrane Fuel Cell (FC).

Supervision design of the system
As discussed before, the supervision platform can be seen as an automaton representing an operating modes management. Three operating modes are distinguished ( Figure 5) OM1: Low power This mode is accessed when the power generated by the renewable sources does not cover the demand. In this case, the batteries are drained at first and then eventually the FC is triggered to use the stored hydrogen as a power back-up and prevent the power shortage. OM2: High power This mode is activated when the power generated by the renewable sources overcomes the required load. The power surplus is then stored as hydrogen using the Electrolyzer or/and as electricity using the batteries. OM3: Safe power This mode is activated when the system fails to provide the required power or when one or more faults occur which make some critical components to be unavailable. The availability of the hardware resources, used in each version, for each mission and then for each mode are evaluated by FDI system to control the transition signal  ij from the mode i to the mode j.

Simulation and Experimental results
The global system has been simulated using dedicated software ©20sim. Two scenario have been considered. Scenario 1 concerns the normal faultless behavior of the system represented by Figure 6. The simulation uses 24 hours weather data of a sunny, average winds day1. The batteries are initially charged at 32% and the hydrogen pressure is about 93% of the maximum capacity of the tank. Between t=[00:00h] and t=[11:45h], the sum of both wind and the solar powers P r were more than enough to satisfy the load P r > 200W . The system starts then in OM 2 : High power. The surplus is stored as hydrogen and electricity using both the electrolyzer and the battery. Between t=[11:45]h and t=[13:02h], the generated power of the sources drops but still more than the load demand 100W < Pr < 200W. The system still operating in OM2, the activation condition of the electrolyzer is then satisfied, as a result, the EL is deactivated and only the batteries are used to store power. After t=[13:02h], the generated power of the sources drops to less than load required power Pr < 100W. As a consequence, the system switches to low power mode OM1.     (Figure 7a) overpasses the thresholds indicating the leak detection. When the detection occurs at t=[10:30]h the hydrogen tank is marked as unavailable, consequently the unique version of the mission: secure the H 2 become unavailable. Consequently, OM1 and OM2 become inaccessible. According to the OMM, the system must switch to the safe mode OM3 (Boolean variable  23 or  13 controlled by online FDI becomes 1) when the H 2 leak occurs, fault has been detected and isolated, both sources (wind and solar) are stopped, and their output powers are zero ( Figure 6). The Figure 7b illustrates the generated and consumed powers of FC and EL. It shows that the EL is also shut-down after the detection, while the FC is activated in order to reduce the H 2 pressure.

Conclusion
A Hybrid Renewable Energy System is a complex mechatronic system because of coupling of several kind of energies. Optimal operating mode management of such process need an unified language for modelling and supervision in terms of safety and control. For those tasks it is shown how the Bond Graph as an unified methodology and a real integrated tool is used not only for modelling but also for supervision design. The developed methodology is applied here online to a real multisource platform.