An Experimental Platform for Monitoring, Identification, and Control of HVAC Processes

An Experimental Platform for Monitoring, Identification, and Control of HVAC processes ⋆ Peter Minarčík ∗,∗∗ Hynek Procházka ∗Martin Gulan ∗∗ ∗Prosystemy, s.r.o., Budmerice 339, 90086 Budmerice, Slovakia (e-mail: [email protected]), ∗∗ Faculty of Mechanical Engineering, Slovak University of Technology in Bratislava, Námestie slobody 17, 812 31 Bratislava, Slovakia (e-mail: [email protected]). Abstract: This paper presents the concept of an intelligent control system platform with advanced control features integrated into a unique laboratory system for simulation of heating, ventilation and air-conditioning (HVAC) processes. The laboratory HVAC system itself constitutes a basis for development and experimental verification of the proposed advanced control algorithms for signal analysis, process identification and controller design, tuning and optimization. Developed advanced control functions for detection of unexpected or undesired signal behaviour, process identification and discrete-time proportional-integral-derivative (PID) and two-degrees-of-freedom (RST) controller tuning are demonstrated. Functionality of the proposed advanced supervision concept integrated into the laboratory HVAC system is demonstrated using preliminary results obtained from practical testing. The concept is built upon an open-source and cost-affordable hardware control platform, and is essentially applicable not only for smart buildings but also various industrial applications. Keywords: HVAC system, control education, smart buildings, monitoring, feedback control 1. INTRODUCTION Smart and green building solutions, along with the Industry 4.0, are nowadays popular and frequented terms. They refer to a general idea, that Heating, Ventilation and Air-Conditioning (HVAC) technologies in buildings, same as production lines in factories, shall be in some way more intelligent. This intelligence is supposed to bring fully automated more cost-effective and more reliable operation with comfortable and intuitive Human–Machine Interface (HMI). In common control practice the ‘smart’ system typically ends up as a standard control system augmented with an additional package of communication protocols, sensors, detectors and actuators. More auxiliary processes are automated and more process data is collected, and displayed in different forms. How to process the collected data remains unclear. We believe that ‘smart’ control engineering must focus on data processing and thus on implementation of unconventional numerical procedures into common practice as well as in teaching curricula. By advanced control techniques we refer to control techniques not commonly used in everyday control engineering ⋆The authors gratefully acknowledge the contribution of the Slovak Research and Development Agency under the grants APVV-18-0023 and APVV-20-0046, the Cultural and Educational Grant Agency of the Ministry of Education of Slovak Republic under the grant 012STU-4/2021. This paper was funded by the European Union under Horizon Europe Grant Agreement number 101079342 (Fostering Opportunities Towards Slovak Excellence in Advanced Control for Smart Industries). practice. This includes model-based control design, robust control, adaptive control, fault detection and fault-tolerant control, model predictive control, etc. All these techniques are well developed and documented, including studies focused on HVAC systems in smart buildings, see e.g. Mirnaghi and Haghighat (2020); Gholamzadehmir et al. (2020); Yao and Shekhar (2021); Afram et al. (2017). However, their use is still mostly limited to universities, research laboratories and R&D centres. Moreover, these techniques are typically applied in specific use cases rather than in common practical control setups, the main reason being their implementation complexity. This paper presents a feasible way towards fulfilling the above-mentioned idea for implementation of advanced control technology. For this purpose an open-source control platform coupled with a novel experimental laboratory HVAC system is used. Note that there are several commercially available physical models providing different possibilities to teach HVAC systems fundamentals as well as to train workers and engineers to operate systems of this type in buildings (we have listed a few of them in Minarčík and Gulan (2021)). Such systems provide many advantages but also disadvantages for the end-users. In order to avoid any compromises, we have created a unique laboratory HVAC system, which is presented in this paper. For control of this experimental setup, different commercial platforms could be used, yet none of them are equipped with satisfactory features, namely with support for advanced control. Therefore, a recently developed open-source control system platform has been utilised – Dynamic Control Unit (DCU) (Procházka, 2015; Prosystemy, 2022), the use of which represents up to two-thirds savings in procurement and operating costs compared to the mentioned competition. Advanced control support means the possibility of free extension of the control platform with new selfdesigned advanced functions, and programming directly in numerical computing platforms (MATLAB or Scilab). This concept of experimental HVAC system is very convenient for teaching purposes as a laboratory trainer, but also for research purposes in cooperation with industry. The rest of paper is structured as follows: Section 2 presents the laboratory HVAC model and its functionality description. Control system platform together with integrated advanced functions, making the control system more versatile and smart is briefly discussed in Section 3. The preliminary results are presented in Section 4. The paper concludes with some final remarks and objectives of future work. 2. LABORATORY HVAC SYSTEM The laboratory HVAC system presented in this section is designed for the simulation of heating, ventilation and airconditioning processes in buildings as well as to validate developed identification, control, and monitoring procedures useful in terms of predictive maintenance, fault or anomaly detection, etc. The first brief introduction to the proposed laboratory HVAC system was in Minarčík and Gulan (2021). Together with its basic description, one of the advanced function for process monitoring was presented. Within this section, the description of the laboratory system hardware will be discussed. As mentioned in Minarčík and Gulan (2021), the laboratory HVAC system had to meet several requirements, which could be summarised as follows: (1) It has to consist of at least one room conditioned with heating, ventilation and air-conditioning circuit capable of supplying/releasing air in the model from/to surroundings; (2) The room should be surrounded at least from one side by so called interspace or outer room representing exterior; (3) The control tasks have to include at least two process values – air temperature and relative humidity inside the object by generating adequate control action to heating/- cooling unit and humidifying device; (4) It has to be able to generate disturbances in terms of medium temperature in mentioned interspace consisting of heating/cooling unit; (5) It must integrate an appropriate human/machine interface to control and monitor the ongoing processes; (6) Modern, practical, robust, and affordable hardware. According to Minarčík and Gulan (2021), the construction of the laboratory model of HVAC system can be divided into three fundamental parts – main or reference room, adjoining room and air ducting. Reference room, henceforth referred to as room-1, represents a particular university classroom, having one neighbouring wall with exterior. Dimensions of room-1 are scaled in 1:10 ratio to real laboratory, hence the inner dimensions are 1075 mm ×625 mm ×323 mm. Thermal comfort conditions are desired to be achieved in this space, while in the adjoining room (room-2) a pseudo-random thermal disturbance will be generated at the same time, in terms of exterior thermal condition simulation. This room has more than 4 times smaller volume with respect to room-1. Its dimensions are 1075 mm×135 mm×323 mm, and it is situated behind room-1 from front view (see Fig. 1, where room-1 is marked by (m) and room-2 by (l) letter). Both rooms are hermetically closed thanks to the characteristic construction. The wall is made of profiles of an aluminium modular system with lining from clean polycarbonate panels inserted in the profile groove with a rubber seal. Polycarbonate wall is covered by expanded polystyrene from both sides despite of the front opening door (Fig. 1 (e)) and dividing wall between room-1 and room-2. The top layer of walls is made of thin plastic plates with printing which is glued onto polystyrene from both interior and exterior sides. Label (d) in Fig. 1 points on air ducting constructed from plastic ducts and fittings covered by polystyrene isolation and clean polycarbonate tubes without any isolation to provide insight into the duct, where sensors and actuators are situated. Air handling duct goes over the whole room1 from one side to the other side of the model, on which two air inlet vents and two air outlet vents are designed. Ductwork includes heating/cooling unit (HCU in further) labelled by (b) to increase/decrease temperature of the air passing through, humidification unit with water reservoir (c) used for increasing the level of relative humidity and air dampers operated by servomotors marked by (a). In order to increase/decrease air temperature in room-2 another HCU (HCU-2) is used, which is directly immersed to the space of room-2. Both of HCUs include two radiators for cooling down the flowing medium. It is radiator marked by (h) for HCU-2 and radiator (i) for HCU-1. Another part of the laboratory model is the electrical control panel (k) equipped with all relevant electronic components and DCU control unit. To monitor and control behaviour of the system, an HMI is required for its operator. This is provided by a touch display and PC peripherals (g). Process variable data are collected from 11 sensors, from which 7 is for temperature sensing only and 4 for temperature as well as relative humidity sensing. Process data recording, web-based visualisation, and advanced computing is provided by server (f), being a standard PC. Derived from three fundamental processes performed by HVAC systems, the laboratory model works in three modes – heating, cooling and ventilation. In the following lines, these operating modes and functionality of the model will be described. Functionality of the operating modes in laboratory HVAC system is based on forcing the air through ducts into room-1, where the thermal comfort is required and controlled. As it has been stated above, HCU-1 and humidifier are used for achieving desired values of air temperature and relative humidity. Together with the ducting dampers, they secure the air flows in suitable trajectory. In order to ensure the most possible interaction between the supplying air and the air in room-1, the ‘Xlogic’ of the air flow forcing is imposed, meaning that the supplying air outgoing from ceiling opening will go out through the ducting grille at the floor level and vice versa. This idea is supported by the fact that the passing air along this path overcomes the longest possible trajectory A C D G E F H JK L M 1 2 3 B I A Fig. 1. Laboratory HVAC system – CAD model in three views (left) and physical model (right). (a) – servomechanisms, (b) – HCU-1, (c) – reservoir with humidifier, (d) – air duct, (e) – opening door, (f) – computer, (g) – HMI with PC peripherals, (h) – cooler for HCU-2, (i) – cooler for HCU-1, (j) – adjustable holder, (k) – control box, (l) – room-2, (m) – room-1. as well as the air will encounter obstacles in room-1, causing a turbulent flow. In the heating mode, the ducting dampers are in a position enabling to exhaust air from room-1 through ducting grille at the ceiling level and the returned air is then sucked by a fan directly to HCU-1 (in heating mode) in the middle of the ductwork, where the air is thermodynamically conditioned by this unit and by a humidifier situated behind HCU-1 (in sense of air flow direction). This conditioned air is subsequently forced by fans at HCU-1 to room-1 through input air grille at the floor level, to which it is directed by air dampers. The supplied air is then expanded into room1 and creates desired thermal conditions in this space. At the same time the air is exhausted from the room and this goes in the cycle while desired temperature and relative humidity of the air are achieved. A similar situation occurs in the cooling mode, where the air circulation has the same direction and is subject to conditioning by HCU-1 (in cooling mode) and the humidification unit if needed. The difference is in a position of ducting dampers, which direct the conditioned air to the inlet vent at the ceiling level of room-1. Supplying air is then distributed into room-1 and exhausted through the outlet ducting grille. The exhausted air is then directed to HCU-1 for further conditioning and the cycle is repeated until desired temperature and relative humidity of air in room-1 are attained. In the ventilation mode it is required to supply the fresh outside air into room-1. This is provided by blowing outside air by ventilator directly through HCU-1 to room-1. HCU-1 is not in heating or cooling mode, only ventilators are operated (ventilation mode). The air is then directed through arbitrary intake vents and supporting fresh air is consequently exhausted from room-1 through arbitrary output openings and it is releasedto exterior as a relief air. Fresh outside air can be also added to the return air during the heating and cooling operation, so the mixed air is brought to HCU-1, and the procedure is repeated. The DCU control platform is used for control and monitoring. It consists of DCU56IO control unit, server and control network. Server is recording collected data to MySQL database and presents data on web-pages. Additionally, the server provides advanced control functionalities, which are discussed in Section 3. 3. DCU CONTROL SYSTEM PLATFORM 3.1 Structure of the DCU Control system platform The DCU control system platform is an open-source control system platform enhanced with educational/research advanced control support. A simplified structure of the platform is in Fig. 2. Blocks represents hardware/software packages, arrows indicates main data flow directions. Right side of the structure depicts classical control system consisting of real-time control units (DCU), process data server and visualisation pages. Left side of the structure is advanced control enhancement of this structure. Real-time control algorithms for DCU control units are prepared in the form of functional diagrams in Simulink/- MATLAB or Xcos/Scilab dynamic simulators. The scientific computational platforms MATLAB/Simulink and Scilab/Xcos are particularly suitable, as they are common tools for advanced control research and development. Designed and if needed simulated control diagrams can be uploaded to DCU control unit (a). DCU collects real-time process data, evaluates uploaded control algorithms and control outputs/actuators accordingly. Selected process variables are sent (d) to the process data server. The server is recording process data to MySQL database and provide them, if requested, to visualisation web-server (g) or to advanced supervision server (e). Advanced supervision server is reading definition of supervision procedures via visualisation web-pages or definition files (f). According to defined procedures process data are requested from data server (e) and they are sent to computational platform while calling defined advanced functions (b). Results from called advanced control functions are sent back to advanced supervision server together with information if the evaluation obtained applicable result. Applicable results are recorded and presented to operator/user via web-pages (f) (in form of alarms, notifications, graphs). In future, the result may be applied directly into control unit (c). a bcd fg e B CD A EGF Fig. 2. Simplified block structure of the DCU Control system platform with main communication directions. Application of advanced control techniques in common control system practice has several requirements. Firstly, the procedure of process data gathering, evaluating advanced techniques, recording and presenting results must be completely automated. Next, advanced techniques must be correctly pre-set in advance or to have some auto-tuning capability so that a common control engineer does not need all knowledge and experience for their reliable use. As stated in Minarčík et al. (2022) activating or setting up an advanced technique must be a matter of a few clicks comprehensible to a common control engineer. Parameters to set, if any, must be related to practical control issues understandable to the control engineer and not to the internal behaviour of the given technique. Presenting results of advanced techniques must be also control engineering oriented. Finally, implementing a new advanced functionalities must be reasonably simple. A researcher/student must be able integrate a new technique easily. Implementation of advanced control techniques into control system platform is supposed to offer to control system engineer a universal analytical tool. A tool for continuous on-line process analysis, identification, controller tuning, fault-detection, while paying attention to the intuitive and simple operation by the end-user. An overlap of aforementioned features and requirements of Prosystemy company resulted in creation of advanced supervision support package integrating advanced functions. These functions can be classified in two classes: (1) Process variable signal analysis – anomaly detection in the signal based on time-domain and frequency-domain analysis, system components’ wearing, detecting unwanted oscillating and switching frequencies, etc.; (2) Process dynamics analysis – model identification and validation, controller auto-tuning based on the identified process, advanced control algorithms – model predictive control, etc. The considered advanced control functions should provide the control system smart and advanced operation in its background. Manager or operator of the system only needs to define several required boundary conditions and values arising from the expected process behaviour. Several of them were programmed, verified and validated, such as the procedure for signal anomalies detection (Minarčík and Gulan, 2021; Minarčík et al., 2021), linear system identification (Minarčík et al., 2022) and a controller tuning procedure. 3.2 Integrated advanced control functions Process signal fault detection function represents the first advanced procedure integrated into the DCU platform. As stated in Li et al. (2020), reliability and accuracy of sensor measurements are crucial for the implementation of advanced control strategies as well as for a correct operation of the system. Therefore, the detection of all possible faults which may cause serious problems or system components’ damage is required to be revealed at its early stage Lughofer et al. (2020). Additionally, internal process control signal analysis may discover problems in control algorithms, for example badly tuned control loops. As mentioned in Majewski and Wojtyna (2017), both timedomain and frequency-domain analysis are very useful to address these problems and they are also widely used for this purpose. Our approach to signal analysis combines the features of three different types of analyses. The first, Fast Fourier transform (FFT) analysis, by converting a signal from time domain into frequency domain. The second, peaks analysis, within which the peaks of the signal are found and the last function provides sudden change analysis, where the sudden change of a signal value is determined. This is provided by a sequential analysis technique named cumulative summation (CUSUM). The signalCheck() is a function used to continuously check selected signals to detect any anomalies and to evaluate possible oscillations above some predefined values of the period, amplitude, and duration. These critical values are defined by user/operator via web-pages. This function can also detect a sudden change in signal amplitude above the user defined threshold. The function consists of reading the input data, processing it and finally writing the results to the output text file. This function was integrated into DCU platform advanced analysis package. The first official introduction of this advanced function was presented in the open-access article Minarčík et al. (2021), in which the procedure was described and supported by simulation and experimental data. However, on the basis of long-term experiments in online operation, minor discrepancies were detected and debugged. Process modelling and identification function is the next implemented advanced control technique. Linear timeinvariant model identification methods are used. They require only process input and output signals to identify a black-box process model. This process modelling approach is the most common technique in many fields including building automation. For instance, Afram and JanabiSharifi (2015) developed in their work black-box models of several subsystems of HVAC system or in Afram et al. (2018) authors developed black-box models of HVAC systems in residential sector. The goal of Royer et al. (2014) was to develop an multi-use black-box modelling procedure for buildings and many others, see e.g. Killian et al. (2015); Guo et al. (2021); Zeng et al. (2021). There are several structures of the transfer function models (black-box models), from which the most common in signal processing applications, ARX (Auto-Regressive with eXternal input) and ARMAX (Auto-Regressive Moving Average with eXtra input), were integrated so far. The function for modelling and linear data-driven identification, named processIdent() was developed. It is an autonomous procedure integrated into DCU platform. Three process signals and several user-defined parameters are required for its execution. The procedure uses black-box modelling and parameter identification of a linear mathematical model in transfer function form. Furthermore, appropriate data pre-processing is also integrated into the identification procedure to increase the procedure reliability. Based on informativeness of the datasets, several model structures are identified, validated and compared. The best model is then automatically selected and compared with the previously identified or initial model. The developed identification function, belongs to the second group of advanced functions – process control dynamics analysis. Analogously to the previous case (signalCheck()), this function has already been introduced as well Minarčík et al. (2022). In this open-access article, we provide a detailed description of the processIdent() procedure. The procedure is still in development and it has been enriched with the estimation of the time-delay (dead time) from input and output data as well as some graphical modifications and another minor improvements. PID and RST controller tuning function is under development as well. The function is supposed to continuously redesign selected PID or RST controllers based on obtained models (from function described above) and to propose successful designs to operator for implementation. Digital PID controller parameters may be computed by different strategies, see e.g. Borase et al. (2021); Bansal et al. (2012); Johnson et al. (2005). The pole placement controller design method of Landau and Zito (2006) was implemented. Its advantages includes that there are no restrictions on the degrees of the discretetime plant model polynomials, there is no restriction on time delay and also on the system stability. The digital controller needs to be defined in canonical three-branched structure (RST), being a general structure of the linear digital controllers. The pole placement method is based on process model and desired closed-loop dynamics defined by required closed-loop poles. The developed pidTune() function, aims to continuously tune RST or digital PID controllers. The process model is obtained by the function processIdent() described above. So to evaluate a new controller, only desired closed-loop poles has to be defined. Definition of the closed-loop poles is automatic an it is based on model dynamics and optimised searching for the best closed-loop response in simulation. User may define if desired controller should be more robust or more aggressive. Output of the developed function are discrete-time PID parameters or RST (two Degrees Of Freedom controller—2-DOF controller) parameters respectively. First and second-order process model structures have been tested for controller design so far. Note that it is sufficient to cover the vast majority of processes in practice. Proposed procedure also check the closed-loop (CL) stability by evaluating the CL roots’ geometrical location in the discrete plane. More practical details about the proposed functions are included in Section 4. 4. PRELIMINARY RESULTS The laboratory HVAC system described in Section 2 was used as a mean for proposed advanced procedures validation in laboratory conditions. The graphical as well as numerical validation results of process signal faultdetection function have already been discussed in Minarčík and Gulan (2021). We therefore present the validation results from two other advanced functions – process model identification and controller tuning. Identification of humidifying process as well as heating and cooling processes are discussed. Three experiments (marked as experiment A, B, and C) were selected and their results are presented. Simplified experimental setups of the user-defined input parameters for evaluating the proposed procedures are captured in Tab. 1. For more information see Minarčík et al. (2021, 2022). Within experiment A we selected three signals: input signal, being the humidifier input power (0 % to 100 %), measured values of relative humidity [%] in room-1 and the signal which enables/disables the identification procedure Table 1. Experimental setups for experiments A—C – input and output structure. Experiment A Experiment B Experiment C Input signal length 5400 samples 20000 samples 20000 samples Output signal length 5400 samples 4000 samples 40000 samples Enable signal length 5400 samples 20000 samples 20000 samples ID 6 100 110 Min. signal length 1200 s 900 s 900 s Max. model order 2 2 2 Max. overshoot 0.02 0.02 0.02 Best fit [%] REAL REAL REAL Model improvement BINARY BINARY BINARY Model polynomials TF TF TF Model delay REAL REAL REAL RST VECTOR VECTOR VECTOR PID VECTOR VECTOR VECTOR The output structure is highlighted by grey colour, where only the data type of parameters are given (numerical outputs from experiments A—C are given in Tab. 2. Note: TF – Transfer Function. (1/0), representing on/off state of the fans of the HCU-1. The input structure contains all input parameters as well as the result structure as shown in Tab 1. The procedure is scheduled for evaluation in pre-defined time period. Numerical results are saved in the database on the local server and the graphical results are exported as well. Exported graphs from experiment A are captured in Fig. 3. The first figure (a) shows signals defined by the user for process identification and their segmentation. It can be observed that one segment has met all requirements. It is entire input signal, entire enable/disable signal and a part of the output signal. This is caused by different sampling time of signals, being 1 s for input and enable/disable signals and 5 s for output signal, thus the output signal is 5 times longer than the other. The second graph (b) captures input and output signals, being step response of the system on two reference changes, which are used for model identification. Identified first order ARX model response shows 92.39 % fit with system response. The open-loop as well as closed-loop step response is shown in the third graph (c). The experiments B and C have the same input parameters setup, but the process model was identified in two different operating modes. Thus, three same signals were selected, being the HCU-1 input power (0 % to 100 %), the average temperature in room-1 [◦C] and the enable/disable signal which is the same to experiment A – operation state of HCU-1’s fans. Figure 4 captures graphical results of the identification procedure of air temperature evaluation over time in the heating operating mode of the laboratory HVAC system, while two step changes were used for excitation of the system. During this procedure, signals with the same length (Fig. 4(a)) were used and the second order ARX model was identified with 93.35 % fit, as shown in Fig. 4(b). When the model was identified, then the controller was computed. It can be observed from Fig. 4(c) that in this case the controller with large difference between CL steady state and reference signal was found. On the other hand, in the cooling operating mode a firstorder ARMAX model with was identified a 85.48 % fit (Fig. 5(b)), and used for computation of controller param0 20 10 30 Detrended input 1.66636e121.666358e12 1.666362e121.666357e12 1.666359e12 1.666361e12 0 10 5 15 Unix time [-] Detrended output 0 2 000 4 0001 000 3 000 5 000500 1 500 2 500 3 500 4 500 0 10 5 15 Validation time [s] Amplitude System response Model response 1. order ARX [[1,-0.9988855,][0,0.0006991,][]83] (92.39%  t) 1.66634e12 1.66636e121.66633e12 1.66635e12 0 20 10 30 -5 5 15 25 Unix time [-] Humidi  er input power [%] 40 60 50 70 45 55 65 75 0 2 -1 1 -0.5 0.5 1.5 RED - input signal; BLUE - output signal; GREEN - enable signal ( * - segment startpoint, o - segment endpoint). Relative humidity [%] 0 10 0002 000 4 000 6 000 8 000 12 000 0 1 0.2 0.4 0.6 0.8 1.2 1.4 Time [kT] Amplitude Reference signal Open-loop response Closed-loop response [r0, r1, r2] = [3.4597579, -3.4553744, 0] and [K, Ti, Td, N] = [3.4553744, 788.28007, 0, 1000] a b c Fig. 3. Results of the evaluation of experiment A. eters and simulation of closed-loop response (Fig. 5(c)). All obtained numerical results are summarised in Table 2. 1.6667e12 1.66672e121.66671e121.666705e12 1.666715e12 0 10 2 4 6 8 12 1 3 5 7 9 11 Unix time [-] HCU-1 input power [%] Air temperature [°C] 22 23 22.2 22.4 22.6 22.8 22.1 22.3 22.5 22.7 22.9 23.1 0 2 -1 1 -0.5 0.5 1.5 RED - input signal; BLUE - output signal; GREEN - enable signal ( * - segment startpoint, o - segment endpoint). 0 10 5 Detrended input 1.6667e12 1.66672e121.66671e121.666705e12 1.666715e12 0 1 0.5 Unix time [-] Detrended output 0 20 00010 0005 000 15 000 0 1 0.5 Validation time [s] Amplitude System response Model response 2. order ARX [[1,-1.9974644,0.9974657,][0,-0.0000062,0.0000064,][]1] (93.35%  t) 0 10 0002 000 4 000 6 000 8 000 12 000 14 000 16 000 18 000 0 1 -0.2 0.2 0.4 0.6 0.8 1.2 Time [kT] Amplitude Reference signal Open-loop response Closed-loop response [r0, r1, r2] = [77.9695, -155.7941, 77.82473] and [K, Ti, Td, N] = [0.14468, 2274.0124, 537.9139, 1000] a b c Fig. 4. Results of the evaluation of experiment B. 5. CONCLUSION This paper presented a novel laboratory HVAC trainer. It emulates a real-world HVAC system equipped with 1.6668e121.66679e121.666785e12 1.666795e12 0 10 2 4 6 8 12 1 3 5 7 9 11 Unix time [-] HCU-1 input power [%] Air temperature [°C] 23 22.2 22.4 22.6 22.8 23.2 23.4 23.6 0 2 -1 1 -0.5 0.5 1.5 RED - input signal; BLUE - output signal; GREEN - enable signal ( * - segment startpoint, o - segment endpoint). 0 10 5 Detrended input 1.6668e121.66679e121.666785e12 1.666795e12 0 -1 -0.5 Unix time [-] Detrended output 0 20 00010 0005 000 15 000 0 -1 -0.5 Validation time [s] Amplitude System response Model response 1. order ARMAX [[1,-0.9998330,][0,-0.0000239,][1,0.0106955,]1] (85.48%  t) 0 20 000 40 000 60 00010 000 30 000 50 000 70 000 0 -1 -1.2 -0.8 -0.6 -0.4 -0.2 -1.1 -0.9 -0.7 -0.5 -0.3 -0.1 Time [kT] Amplitude Reference signal Open-loop response Closed-loop response [r0, r1, r2] = [-1.38094, 1.38052, 0] and [K, Ti, Td, N] = [-1.38048, 3299.4488, 0, 1000] a b c Fig. 5. Results of the evaluation of experiment C. a full-scale control platform that additionally supports advanced control techniques. In consequence, the system Table 2. Numerical results of the experiments. Experiment A Experiment B Experiment C [1,-0.998886] [1,-1.9974644,0.9974657] [1,-0.9998330] Model [0,0.000699] [0,-0.0000062,-0.0000064] [0,-0.0000239] [1,0.0106955] Delay 83 samples 1 sample 1 sample FIT 92.39% 93.35% 85.48% 3.4597579 77.9695 -1.38094 RST -3.4553744 -155.7941 1.38052 0 77.8247 0 3.4553744 0.14468 -1.38048 PID 788.28007 2274.0124 3299.4488 0 537.9139 0 The first line of model section corresponds to coefficients of deterministic and stochastic model denominator, the second to deterministic model numerator and the third to the stochastic model numerator. Analogously in RST and PID sections, where the lines correspond to controller parameters (r0,r1,r2for RST and K,Ti,Tdfor PID). can be used for a variety of educational and research purposes. Students are able to learn standard control engineering practices of programming or maintaining a building control system while observing the underlying control system theory. Researchers may use the system for simulation, process identification or design of advanced control techniques. Some of the possible tasks for students and researchers that can be performed on such a testbed could be summarised as follows: (1) simulate and study the fundamentals of HVAC processes; (2) develop, verify and/or validate different types of control and identification procedures for HVAC processes; (3) obtain experimental data for the purposes of offline training and simulation associated with the teaching process; (4) graphical programming in the MATLAB/Simulink or Scilab/Xcos environment with the possibility of online verification of the proposed solution; (5) detection and elimination of artificially created faults, etc. We provided a description of the physical model, its main components and demonstrated its functionality. The laboratory system was used for experimental validation of proposed advanced control functions. Via experiments we demonstrated the procedures of system identification and tuning of discrete PID and RST controllers, and presented their graphical and numerical outputs. From the results of the experimental validation, it is possible to state that the proposed procedures have a correct and satisfying functionality. REFERENCES Afram, A., Fung, A.S., Janabi-Sharifi, F., and Raahemifar, K. (2018). Development and performance comparison of low-order blackbox models for a residential HVAC system. Journal of Building Engineering, 15, 137–155. Afram, A. and Janabi-Sharifi, F. (2015). Black-box modeling of residential HVAC system and comparison of gray-box and blackbox modeling methods. Energy and Buildings, 94, 121–149. Afram, A., Janabi-Sharifi, F., Fung, A.S., and Raahemifar, K. (2017). Artificial neural network (ANN) based model predictive control (MPC) and optimization of HVAC systems: A state of the art review and case study of a residential HVAC system. Energy and Buildings, 141, 96–113. Bansal, H., Sharma, R., and Ponpathirkoottam, S. (2012). Pid controller tuning techniques: A review. 2, 168–176. Borase, R., Maghade, D., Sondkar, S., and Pawar, S. (2021). A review of PID control, tuning methods and applications. International Journal of Dynamics and Control, 9. Gholamzadehmir, M., Del Pero, C., Buffa, S., Fedrizzi, R., and Aste, N. (2020). Adaptive-predictive control strategy for HVAC systems in smart buildings – a review. Sustainable Cities and Society, 63, 102480. Guo, Z., Coffman, A.R., Munk, J., Im, P., Kuruganti, T., and Barooah, P. (2021). Aggregation and data driven identification of building thermal dynamic model and unmeasured disturbance. Energy and Buildings, 231, 110500. Johnson, M. et al. (2005). PID control: New Identification and Design Methods. Springer. Killian, M., Mayer, B., and Kozek, M. (2015). Effective fuzzy black-box modeling for building heating dynamics. Energy and Buildings, 96, 175–186. Landau, D.I. and Zito, G. (2006). Digital Control Systems: Design, Identification and Implementation. Springer, 1 edition. Li, D., Cai, Z., Qin, B., and Deng, L. (2020). Signal frequency domain analysis and sensor fault diagnosis based on artificial intelligence. Computer Communications, 160, 71–80. Lughofer, E., Zavoianu, A.C., Pollak, R., Pratama, M., MeyerHeye, P., Zörrer, H., Eitzinger, C., and Radauer, T. (2020). Online anomaly detection with advanced independent component analysis of multi-variate residual signals from causal relation networks. Information Sciences, 537, 425–451. Majewski, J. and Wojtyna, R. (2017). Results of applying evolutionary algorithms to frequency-domain signal analysis. In Signal Processing: Algorithms, Architectures, Arrangements, and Applications, 121–124. Minarčík, P., Procházka, H., and Gulan, M. (2021). Advanced supervision of smart buildings using a novel open-source control platform. Sensors, 21(1). Minarčík, P., Procházka, H., and Gulan, M. (2022). A data-driven identification procedure for HVAC processes with laboratory and real-world validation. Processes, 10(1). Minarčík, P. and Gulan, M. (2021). Intelligent supervision concept integrated into a laboratory HVAC system. In 23rd International Conference on Process Control, 319–324. Mirnaghi, M.S. and Haghighat, F. (2020). Fault detection and diagnosis of large-scale HVAC systems in buildings using datadriven methods: A comprehensive review. Energy and Buildings, 229, 110492. Procházka, H. (2015). Development of universal control platform a new advanced fully featured control system for general public. Prosystemy, s. (2022). About us. URL http://prosystemy.sk/ index.php?lang=en. Royer, S., Thil, S., Talbert, T., and Polit, M. (2014). Black-box modeling of buildings thermal behavior using system identification. IFAC Proceedings Volumes, 47(3), 10850–10855. Yao, Y. and Shekhar, D.K. (2021). State of the art review on model predictive control (MPC) in heating ventilation and airconditioning (HVAC) field. Building and Environment, 200, 107952. Zeng, T., Brooks, J., and Barooah, P. (2021). Simultaneous identification of linear building dynamic model and disturbance using sparsity-promoting optimization. Automatica, 129, 109631.