BIMERR (BIM-based holistic tools for Energy-driven Renovation of existing Residences) is a research and innovation program which started in 2018, supported by a consortium of 16 members, including EU companies and Universities, and co-funded by the European Union Horizon 2020 [1
]. The target of the BIMERR project is to design and develop a renovation toolkit which will comprise tools to support renovation stakeholders, throughout the renovation process of existing buildings, from project conception to delivery. It comprises tools for the automated creation of enhanced Building Information Modeling (BIM), a renovation decision support system to aid the designer in exploring available renovation options, as well as a process management tool that will optimize the design and on-site construction process toward optimal coordination and minimization of renovation time and cost. At the heart of the BIMERR toolkit lies an Interoperability Framework which enforces semantic interoperability among BIMERR tools, as well as with third-party legacy Information Communication Technologies (ICTs) tools to enable seamless BIM creation and information exchange among Architecture, Engineering and Construction (AEC) stakeholders in an effort to enhance the rapid adoption of BIM in renovation of the existing EU building stock [2
]. In this study, the BIMERR toolkit was validated and demonstrated in four buildings in three European Member States. Two buildings were used for pre-validation and implementation refinement, and the refined BIMERR toolkit supported the actual renovation design and works in the other two pilot buildings. The assessment and evaluation of the BIMERR toolkit after these real-life activities fed material into two supporting horizontal project activities: (i) dissemination and exploitation of project outcomes through the creation of best practice examples of BIMERR use that will guide further replication effort, and (ii) promotion of BIMERR outcome to the most relevant standardization bodies.
The availability and use of large volumes of BIM-generated data to gain knowledge that can be further exploited in various applications [4
] in a BIM-based project by using machine learning (ML) and deep learning (DL) is a new research area of particular interest for the AEC industry [5
]. BIM models for each stage can be developed from different data sources (e.g., designer inputs, CAD as-build, etc.), and then relevant data can be extracted from the BIM model in order make adaptive/smart decisions.
Early ML-based BIM research dealing with how the BIM process enhances the project-progress tracking was proposed in Reference [7
]. Work in References [8
] integrated cloud BIM, IoT and Data Mining (DM) into the digital twin framework, establishing a high degree of automation. Combining BIM and expert systems in order to provide a way to conduct autonomous safety risk identification is examined in Reference [11
]. Predictive maintenance based on BIM, IoT and ML techniques to improve the facility management of existing building assets is examined in Reference [12
The usability of dynamic monitoring and control for Building Energy Management Systems is presented in Reference [13
] by integrating facilities, sensors and ML combined with simulation models, historical data and an intelligent controller. In Reference [14
] is presented a BIM-based DM method that analyzes thermal performance of buildings with the help of BIM data; Heating, Ventilating and Air-Conditioning (HVAC); environmental conditions; and occupant behavior.
An extensive literature review on machine-learning applications to BIM is out of the scope of this paper; hence, the interested reader is referred to the excellent work presented in Reference [15
The accuracy of the BIMERR energy Key Performance Indicator (KPI) values is highly affected by the level of detail of the input data, and recent studies have shown that the occupant-behavior data constitute the major cause of uncertainty [16
]. Hence, a deeper understanding and proper modeling of the occupant behavior are of paramount importance. For capturing information that goes beyond static data of BIM, the BIMERR toolkit requires specific dynamic sensor data in order to generate occupant behavior (OB) models/profiles, provided in a standardized format (occupant behavior standardized XML-obXML) [17
]. To this direction, the sensor data streams must be sourced from appropriate sensors and able to provide measurements for various parameters of interest, such as room air temperature and humidity, room occupancy, air conditioner’s status (on/off), air conditioner’s thermostat set-point temperature and operation mode (heating, cooling, fan, etc.), status of doors and windows (open/closed), weather conditions (including outdoor temperature and relative humidity, and wind speed and direction) and energy consumption.
The Wireless Sensor Network (WSN) is a suitable solution that is able to integrate the sensing and controlling of the environment in the context of the BIM and smart buildings. Some of the recent studies present different solutions as to how to design WSNs with regard to sensing coverage, measurements significance, network connectivity and spatial positioning of sensors. In Reference [18
], the authors used genetic algorithms to improve WSNs’ location accuracy; in Reference [19
], the authors proposed an evolutionary algorithm to calculate the maximum sensor coverage; in Reference [20
], the authors proposed an information service and live sensing in order to supervise the building energy performance; and in Reference [21
], the authors investigated how BIM could support smart buildings from the architecture plan to the building management and [22
] demonstrated sensor data visualization in BIM for supporting complex decisions.
Applying machine-learning algorithms on the sensor-data streams provided by the WSNs that are designed for and installed in the pilot sites generates occupant-behavior profiles that mimic the inhabitants’ actions and contribute to the accuracy of energy performance estimation, using ML-enabled occupant behavior methods [23
]. These profiles are subsequently used to project the building system (e.g., heating/cooling) utilization boundaries to maintain the comfort zone of the residents.
This paper addresses the deployment, technology, topology and cost of robust WSNs that are able to collect and transmit real-time data required from BIM renovation toolkits. It is based on real WSN installations, rather than simulated results, as in most cases reported in the open literature. As a consequence, it considers different field requirements, constraints and options at the BIMERR pilot sites, and, hence, it sets the lines for similar installations required by the construction industry for collecting dynamic data required for BIM and smart buildings.
The paper is organized as follows: Section 2
describes the principles of the Wireless Sensor Network (WSN) deployed in the pilot sites of the BIMERR project. Section 3
presents the topology of the BIMERR WSN. Section 4
provides details about the equipment, sensors and budget. Section 5
presents measured sensor data from the BIMERR pilots, and Section 6
describes the application of WSN data in the context of BIM for training occupant behavior models.
5. Sensor Data from the Pilots
As detailed in Reference [29
], the Middleware provides identity and data management services to facilitate a secure and resource-friendly flow of information among Gateways and other BIMERR applications. The offered services can be deployed on the cloud or edge inside buildings or apartments. The Middleware design is a modular subsystem customized to satisfy the BIMERR use cases, and it consists of the following components:
Identity Provider: a central authentication server, a directory of user and application profiles, providing interfaces to authenticate them.
Registry: a central repository containing metadata of Gateways and sensors.
Data Processor: retrieves device metadata from Registry and automatically configures the available workflows for data routing and transformation.
Storage: sensor data archival and retrieval.
OTA Software Update and Monitoring: remote configuration and monitoring of the Gateway software.
Device Connector: interoperability between Z-Wave sensors and data processor.
The Middleware acts as the sensor-data handler and repository, and it provides the Registry that contains metadata to query and access required items. The Middleware’s sensor-data storage services are microservices responsible for the storage and orchestration of sensor measurements data across the edge (building location) and BIMERR cloud.
As already mentioned, the accuracy of energy performance, as estimated by building energy performance simulation engines, such as EnergyPlus [30
], is highly affected by the level of detail of its input data. Methods and models have been developed [19
] to reduce the gap between simulated and measured building energy performance by representing occupant behavior in a standardized XML schema (obXML) [17
]. The BIMERR PRUBS (Profiling Residents Usage Building Systems) component [28
] applies machine-learning algorithms on sensor data streams to generate occupant-behavior profiles that mimic the inhabitants’ actions. These profiles are subsequently used to project the building system (e.g., heating/cooling) utilization boundaries to maintain the comfort zone of the residents. A high-level architecture of WSN, Middleware, PRUBS and BIMERR interoperability framework is illustrated in Figure 8
For the occupant-behavior model training, sensing and metering data are acquired by the WSN designed for and installed at each site. Weather data are received from external weather APIs. To receive the sensor data streams, PRUBS communicates with the BIMERR Middleware, which is responsible for the buildings’ sensor-data handling. The PRUBS output contains data-driven occupant-behavior models and populates a building level obXML file. That file is sent to the BIMERR Interoperability Framework [32
] to be properly enriched and linked with other data, so that it can be used by other BIMERR applications. To conceptually gather the data mapping rules that must be performed to transform sensor data that are available through Middleware, and to input data that are meaningful for PRUBS, a Middleware-to-PRUBS sensor data wrapper was developed [28
A demonstration of this wrapper is hereafter presented for the Greek pre-validation site and the two validation sites, i.e., the Spanish and Polish pilot. Figure 9
depicts the temperature profile and air-conditioning (AC) units’ power-consumption levels, as these were measured in the master bedroom of the pre-validation site and captured by the sensor data mapping. WSN data were retrieved for the period from 15 September to 25 September. As it can be seen, the temperature distribution fluctuates in response to the AC usage. In this case, the higher temperature values (over 25 °C) are well-captured during the high-power consumption days (indicated by the AC power spikes), while these are maintained at about 24–25 °C during the days that the power consumption remains relatively low.
Concerning the Spanish validation site, Figure 10
presents temperature data obtained by different sensors within the same apartment. Similar to the Greek pre-validation site, the obtained data indicate the seamless sensor-data flow for the period 5 November 2021–15 November 2021, during which the room temperature adapted in accordance with the occupants’ heating needs. Figure 11
demonstrates, respectively, the total power consumption measured for the whole apartment. An example of the sensor-temperature-data mapping for the second validation site (in Poland) is also shown in Figure 12
for the same period.
The aforementioned sensor data streams are subsequently used by the BIMERR toolkit to train the OB models, as discussed in the following section.
6. A Use-Case Example of WSN Measured Data for BIM: Training Occupant Behavior Models to Increase the Energy-Performance Estimation Accuracy
The methods applied to train occupant behavior models by PRUBS fall under three main categories: (1) the occupant thermal comfort, (2) the HVAC systems’ thermostat adjustment and (3) the profiling of other electric components’ usage. A brief description is provided in the following sections; meanwhile, for a detailed description, the reader should refer to Reference [28
For the occupant thermal comfort modeling, the Gaussian Naïve Bayes method [33
] is applied, a supervised machine-learning algorithm that follows the Bayes theorem, assuming conditional independence between the features, named Drivers in terms of obXML. Before training the model, an event-generation process is applied to populate the proper training events/dataset required as input to the ML model. In that process, certain preprocessing rules/steps are applied to sensed data that are stored in the PRUBS database. It is worth mentioning that other methods have also been evaluated as candidate alternatives to the Gaussian Naïve Bayes classifier for the thermal-comfort profiling, such as the Multinomial Naive Bayes, Logistic Regression, Random Forest, MLP Classifier (Neural Networks), SGD Classifier (Stochastic Gradient Descent), SVM (Support Vector Machines) and K-Neighbors Classification, [33
]. A comparison was performed; however, such an analysis is beyond the scope of this paper.
For the HVAC systems’ thermostat-adjustment behaviors, two statistical modeling approaches that have been introduced in recent studies and seem to be the most applicable to residential buildings [23
] are implemented in PRUBS. In the first approach [23
], occupants are clustered into active, medium and passive users and multivariate logistic regression models are trained to predict the likelihood of a set-point increase or decrease. In the second approach [24
], a three-parameter discrete Weibull distribution model is being developed and trained to represent occupants air-conditioning usage, considering indoor temperature, CO2
concentration and occupancy state as the model Drivers.
The Gaussian Processes [25
] modeling approach was selected to extract information for the Other Electric Equipment Usage. Other Electric Equipment Usage includes all the electric components with a consumption that is not directly measured by any metering device but can be implicitly estimated by subtracting the electric HVAC loads consumption (measured) from the total electric consumption (measured) of the apartment. The Gaussian Processes (GPs) approach is a supervised machine-learning approach that predicts the mean value and standard deviation of the output (“Actions” in terms of obXML) given a set of inputs (“Drivers” in terms of obXML). This machine-learning modeling approach was selected because of its non-parametric nature, since it allows us to use the same models in all the experimental setups, regardless of the specifics of the HVAC systems, whereas, in the case of parametric models, a laborious manual model selection process has to be performed. Concerning its implementation to PRUBS, a trained Gaussian-Processes-based model predicts the mean and standard deviation of the Other-Electric-Equipment-Usage power consumption for each apartment at each time-step, given as input (“Drivers”) the following measured parameters: (1) hourly outdoor air temperatures in °C, relative humidity in %, and cloud coverage for the building’s location received from external weather APIs and stored in PRUBS private repository; (2) rooms’ (space) air temperature, the hourly air temperature for each space of the apartment received from the Middleware and mapped to the PRUBS data model; (3) event driven external windows and doors state data acquired from the Middleware mapped to the PRUBS data model and interpolated to provide hourly data; (4) hourly HVAC load-type power-consumption data received from the Middleware and aggregated to get the hourly HVAC total power consumption; (5) hourly total electric consumption of the apartment received from the Middleware; and (6) time-of-day and day-of-week data.
The performance evaluation of the trained models is based on the FIT
metric, which is defined as follows:
is the normalized root mean square error,
the predicted value,
the measured value and l
is the number of training-set samples.
As an application example, Figure 13
shows for the Greek pre-validation site a snapshot of the populated occupant behavior model, as a result of the corresponding WSN data. Note that the first bunch of devices in the Greek building was on-boarded in October 2020, but, due to the pandemic and the fact that the building was unoccupied from October 2020 to March 2021, the lack of actual HVAC and other electric-equipment power-consumption data for that period made the evaluation of the trained algorithms biased. Nevertheless, they did reflect the actual status of the building and, in this case, co-simulation of the OB models with building energy-performance simulations estimated zero heating demands and thermal comfort boundaries that were significantly lower than the usual temperature set-point limits. Since April 2021, more reasonable sensor data have been collected, and during the summer months, the first non-zero HVAC power-consumption data became available. Based on the trained OB models for the aforementioned period, the corresponding file of Greek building was successfully updated.
Conceptualization, D.K. and G.A.; methodology, D.K., G.T. and G.G.; formal analysis, D.K., G.T., G.G.; writing—original draft preparation, D.K., G.T. and G.G.; writing—review and editing, G.A., G.T. and G.G. All authors have read and agreed to the published version of the manuscript.
The work in this paper was co-funded by the European Union Horizon 2020 Research and Innovation Programme, under grant agreement no. 820621.
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
The authors would like to thank all the BIMERR partners and, in particular, Budimex, Ferrovial, Conkat, FIT and Suite 5.
Conflicts of Interest
The authors declare no conflict of interest.
- BIM-Based Holistic Tools for Energy-Driven Renovation of Existing Residences—BIMERR. Available online: https://bimerr.eu/ (accessed on 1 December 2021).
- EU BIM Task Group. Available online: http://www.eubim.eu/ (accessed on 1 December 2021).
- National Building Specification. Working towards a Unified Approach to BIM in Europe. Available online: https://www.thenbs.com/knowledge/working-towards-a-unified-approach-to-bim-in-europe (accessed on 1 December 2021).
- Krijnen, T.; Tamke, M. Assessing Implicit Knowledge in BIM Models with Machine Learning; Springer International Publishing: Cham, Switzerland, 2015; pp. 397–406. [Google Scholar]
- Dreyfus, H.; Dreyfus, S.E.; Athanasiou, T. Mind over Machine; Free Press: New York, NY, USA, 1986. [Google Scholar]
- Hong, T.; Wang, Z.; Luo, X.; Zhang, W. State-of-the-art on research and applications of machine learning in the building life cycle. Energy Build. 2020, 212, 109831. [Google Scholar] [CrossRef][Green Version]
- Golparvar-Fard, M.; Savarese, S.; Mora, F.P. Automated model-based recognition of progress using daily construction photographs and IFC-based 4D models. In Proceedings of the Construction Research Congress 2010: Innovation for Reshaping Construction Practice—Proceedings of the 2010 Construction Research Congress, Banff, AB, Canada, 8–10 May 2010; pp. 51–60. [Google Scholar]
- Pan, Y.; Zhang, L. Automated process discovery from event logs in BIM construction projects. Autom. Constr. 2021, 127, 103713. [Google Scholar] [CrossRef]
- Pan, Y.; Zhang, L. A BIM-data mining integrated digital twin framework for advanced project management. Autom. Constr. 2021, 124, 103564. [Google Scholar] [CrossRef]
- Zhang, L.; Wu, X.; Ding, L.; Skibniewski, M.; Lu, Y. Bim-Based Risk Identification System in Tunnel Construction. J. Civ. Eng. Manag. 2016, 22, 529–539. [Google Scholar] [CrossRef]
- Cheng, J.C.; Chen, W.; Chen, K.; Wang, Q. Data-driven predictive maintenance planning framework for MEP components based on BIM and IoT using machine learning algorithms. Autom. Constr. 2020, 112, 103087. [Google Scholar] [CrossRef]
- McGlinn, K.; Yuce, B.; Wicaksono, H.; Howell, S.; Rezgui, Y. Usability evaluation of a web-based tool for supporting holistic building energy management. Autom. Constr. 2017, 84, 154–165. [Google Scholar] [CrossRef][Green Version]
- Garcia, L.C.; Kamsu-Foguem, B. BIM-oriented data mining for thermal performance of prefabricated buildings. Ecol. Inform. 2019, 51, 61–72. [Google Scholar] [CrossRef][Green Version]
- Liu, Z.; Zhang, A.; Wang, W. A Framework for an Indoor Safety Management System Based on Digital Twin. Sensors 2020, 20, 5771. [Google Scholar] [CrossRef] [PubMed]
- Zabin, A.; González, V.A.; Zou, Y.; Amor, R. Applications of machine learning to BIM: A systematic literature review. Adv. Eng. Inform. 2021, 51, 101474. [Google Scholar] [CrossRef]
- D’Oca, S.; Fabi, V.; Corgnati, S.P.; Andersen, R.K. Effect of thermostat and window opening occupant behavior models on energy use in homes. Build. Simul. 2014, 7, 683–694. [Google Scholar] [CrossRef]
- Hong, T.; D’Oca, S.; Taylor-Lange, S.C.; Turner, W.J.; Chen, Y.; Corgnati, S.P. An ontology to represent energy-related occupant behavior in buildings. Part II: Implementation of the DNAS framework using an XML schema. Build. Environ. 2015, 94, 196–205. [Google Scholar] [CrossRef][Green Version]
- Zhou, H.; Hu, J.; Wu, H.; Guo, C. Indoor positioning research based on wireless sensor network topology optimization. In Proceedings of the IEEE 2019 Chinese Automation Congress (CAC), Hangzhou, China, 22–24 November 2019. [Google Scholar]
- Yoon, Y.; Kim, Y. An Efficient Genetic Algorithm for Maximum Coverage Deployment in Wireless Sensor Networks. IEEE Trans. Cybern. 2013, 43, 1473–1483. [Google Scholar] [CrossRef] [PubMed]
- Wang, H.; Gluhak, A.; Meissner, S.; Tafazolli, R. Integration of BIM and Live Sensing Information to Monitor Building Energy Performance. In Proceedings of the 30th CIB W78 International Conference, Beijing, China, 9–12 October 2013. [Google Scholar]
- Zhang, J.; Seet, B.-C.; Lie, T.T. Building Information Modelling for Smart Built Environments. Buildings 2015, 5, 100–115. [Google Scholar] [CrossRef][Green Version]
- Chang, K.-M.; Dzeng, R.-J.; Wu, Y.-J. An Automated IoT Visualization BIM Platform for Decision Support in Facilities Management. Appl. Sci. 2018, 8, 1086. [Google Scholar] [CrossRef][Green Version]
- Yan, D.; O’Brien, W.; Hong, T.; Feng, X.; Gunay, H.B.; Tahmasebi, F.; Mahdavi, A. Occupant behavior modeling for building performance simulation: Current state and future challenges. Energy Build. 2015, 107, 264–278. [Google Scholar] [CrossRef][Green Version]
- Gaetani, I.; Hoes, P.J.; Hensen, J.L. Ocupant behavior in building energy simulation: Towards a fit-for-purpose modeling strategy. Energy Build. 2016, 121, 188–204. [Google Scholar] [CrossRef]
- Amasyali, K.; El-Gohary, N. Machine learning for occupant-behavior-sensitive cooling energy consumption prediction in office buildings. Renew. Sustain. Energy Rev. 2021, 142, 110714. [Google Scholar] [CrossRef]
- Jia, M.; Srinivasan, R.S.; Raheem, A. From occupancy to occupant behavior: An analytical survey of data acquisition technologies, modeling methodologies and simulation coupling mechanisms for building energy efficiency. Renew. Sustain. Energy Rev. 2017, 68, 525–540. [Google Scholar] [CrossRef]
- Recommendation ITU-T, G.9959: Short Range Narrow-Band Digital Radiocommunication Transceivers—PHY and MAC Layer Specifications (Z-Wave). Available online: https://www.itu.int/rec/T-REC-G.9959-201501-I/en (accessed on 26 January 2021).
- BIMERR Deliverable D5.8 ‘Building Resident Energy-Related Behavior Profiling Framework 2’. 2020. Available online: https://bimerr.eu/ (accessed on 26 January 2021).
- BIMERR Deliverable D3.6 ‘BIMERR System Architecture 2nd version’. 2021. Available online: https://bimerr.eu/ (accessed on 26 January 2021).
- Crawley, D.B.; Lawrie, L.K.; Winkelmann, F.C.; Buhl, W.; Huang, Y.; Pedersen, C.O.; Strand, R.K.; Liesen, R.J.; Fisher, D.E.; Witte, M.J.; et al. EnergyPlus: Creating a new-generation building energy simulation program. Energy Build. 2001, 33, 319–331. [Google Scholar] [CrossRef]
- EBC Annex 66. Final Report-Definition and Simulation of Occupant Behavior in Buildings. 2018. Available online: http://annex66.org/sites/default/files/2018FinalReport/Annex%2066%20Final%20Report%20-%20read.pdf (accessed on 26 January 2021).
- BIMERR Deliverable D4.9. Integrated BIMERR Interoperability Framework 2. 2021. Available online: https://ploutos-h2020.eu/wp-content/uploads/2021/09/Ploutos_D4.4-Initial-agri-food-data-sharing-reference-architecture.pdf (accessed on 26 January 2021).
- Rasmussen, C.E.; Williams, C.K. Gaussian Processes for Machine Learning; The MIT Press: Cambridge, MA, USA, 2006; Volume 38, pp. 715–719. [Google Scholar]
BIMERR WSN concept.
Z-Wave mesh network concept.
WSN Architecture 1.
WSN Architecture 2.
WSN topology of the pilot apartments in Poland.
WSN topology of the pilot apartments in Spain.
WSN topology of the pilot apartments in Greece.
BIMERR WSN, Middleware, PRUBS and Interoperability Framework architecture outline [28
Greek pilot site’s temperature sensor (left axis—purple color) and AC units power (right axis—green color) consumption data, 15 September 2021–25 September 2021.
Spanish pilot site’s temperature data from different sensors from the same apartment, 5 November 2021–15 November 2021.
Spanish-pilot-site apartment’s total power consumption data, 5 November 2021–15 November 2021.
Polish pilot site’s temperature data from different sensors in the same apartment, 5 November 2021–15 November 2021.
Greek pilot site’s trained thermal comfort bounds and thermostat model.
Polish pilot’s Z-Wave sensors per apartment.
|Sensor Type||Sensor Model||Manufacturer||Quantity|
Polish pilot’s communication and data-processing equipment for all apartments.
|Sensor Type||Sensor Model||Mfn||Quantity|
|Raspberry||Pi 4 Model B 4 GB||RASPBERRY||1|
Spanish pilot’s Z-Wave sensors per apartment.
|Sensor Type||Sensor Model||Mfn||Quantity|
|Wall Plugs||FGWPF-102||FIBARO||1–5 a|
Spanish pilot’s communication and data-processing equipment per apartment.
|Sensor Type||Sensor Model||Mfn||Quantity|
|Raspberry||Pi 4 Model B 2 GB||RASPBERRY||1|
Greek pilot’s Z-Wave sensors per apartment.
|Sensor Type||Sensor Model||Mfn||Quantity|
Greek pilot’s data-processing equipment.
|Sensor Type||Sensor Model||Mfn||Quantity|
|Raspberry||Pi 4 Model B 2 GB||RASPBERRY||1|
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).