BIM is “a digital representation of physical and functional characteristics of a facility” (1). This digital representation (3d model) is rich in data and information which can be shared for multiple purposes, such as energy simulation, clash detection, cost evaluation, and facility operations and management. There are intrinsic synergies between BIM and FM (2). One of the biggest potential contributions of BIM for FM, however, lies in its long-term competence of providing FM services in the operation and maintenance (O&M) phase. The duration of O&M phase spans 30 years or even more, and the cost of building O&M makes up 75%-85% of the total cost, which imposes considerable influence on and offering possibilities of huge cost saving to the industry (3).

One of the main purposes of sensor analytics is to find trends and patterns. By studying these, researchers can not only figure out what happened and is happening, but also predict the change of data in the future. Another purpose is to discover anomalies. By examining deviations from a pre-established setpoint, people have a better understanding of what is not in good condition or out of control. This is significant for proactively preventing equipment failure, warning information can be generated for troubleshooting and maintenance when a part of heating, ventilation and air conditioning (HVAC) system does not function properly (4).

With the advent of “smart building” technology, more buildings have been equipped with intelligent building automation systems (BAS). In a BAS system, several types of sensors are needed to continuously acquire a large amount of data (5). Real-time sensor data helps with evaluating building performance and data-driven decisions related to facility operation and management (6). The employment of BIM-and-sensor-data integration is not confined to buildings. It also applies to other structures that require daily monitoring and management, such as bridges and dams.

One of the strategies to integrate BIM and sensor data is to attach the real-time data to geometric and spatial information extracted from the model (6). Such a “dynamic” BIM goes beyond a static source of information that comprises only limited design and construction data. Through this dynamic BIM, facility managers are able to visualize and monitor the status of sensors and facilities (7). Problems can also be detected and located immediately. Dynamic models have the potential to minimize the cost on facility operation and lowering the possibility of abnormal conditions that jeopardize the comfort and safety of building occupants.

The purpose of sensor fault detection is to identify faults and pinpoint the location and type of them. The definition of “faulty” can be described as the deviation from normal sensor measurements, which indicates an abnormal working status of sensors due to hardware malfunction or failure. However, there is another interpretation of “faulty” based on the concept of thermal comfort. In the context of indoor thermal comfort, “faults” are detected on the basis of temperature and humidity, standards that are set by the ASHRAE and ISO. For example, RH data from a data feed is examined with the comfort RH range (20%-80%) defined by ASHRAE 55-2017. Any RH values lower than 20% or higher than 80% will be defined as “faulty” and not desired for good thermal comfort. Types of “faults,” when defined with a comfort zone in a psychrometric chart, are linked with how people feel about the thermal condition in the space where they are. In this context, there can be up to eight types of faults as combinations of temperature and humidity: “hot,” “dry,” “cold,” humid,” “hot and dry,” “hot and humid,” “cold and humid,” and “cold and dry.”

Building information model can be used is to visually track sensor data and provide fault detection for comfort ranges (temperature and humidity). “Values out of ranges” indicates the existence of anomalous sensor data values. The definition of “ranges” is based on ASHRAE 55-2017 comfort zone, in which temperature ranges from 68ºF to 78ºF, and RH ranges from 20% to 80%. ISO 7730 can also be used for defining ranges.

The value of BIM application in FM lies in that BIM improves the current manual process of information handover and contributes to more accurate and accessible FM data for more efficient work order execution. However, BIM for FM faces challenges ranging from methodologies, shortage of knowledge about implementation requirements, and BIM expertise in FM industry (8). Three different universities showed how BIM could be used to simplify FM workflow (9), integrate an integrated multi-view visualizer for interfacing HVAC information to support troubleshooting (9), and use a third party software EcoDomus as the center of the BIM-FM framework, integrating information from BIM and other systems (10).

Analyzing sensor data alone does not contribute much to understanding facility condition, since data makes far more sense when they are analyzed in a “context” (5). This context can be provided by spatial information from building information models, which enables a visualized and integrated way of data analysis. For instance, by integrating data into BIM from temperature sensors installed in HVAC systems, FM managers can look for discrepancies between temperatures in adjacent rooms to detect system faults (11).

Attempts of integration BIM with various sensing technologies and devices are made in past research projects regarding multiple areas in building automation (12). For example, temperature, humidity, and energy consumption sensors were used to implement post occupancy evaluation and real-time energy performance in a residential building (13). Data retrieved from fire sensors were utilized for establishing a fire alarm management system to determine the authenticity of fire alarms and to prevent disturbance because of false alarms (14).

The integration of sensor data with BIM is not confined to indoor environments shown by the application of RFID and GPS sensors for creating a model that estimates locations of tools, equipment, and materials in construction projects (15) or load sensors in a crane navigation system where the position of lifted objects is displayed with video cameras in the context of a building and surroundings (16).

The structure of sensor readings can vary, but key attributes such as sensor ID, reading ID, data value, and time stamps are always required. Sensor metadata can be considered as descriptions of data, or “data about data” (17). They provide necessary information for sensor discovery and data processing (5). One major purpose of the sensor metadata is to easily keep track of anomalies. For example, when faulty values are detected in data analysis, researchers want to find out which specific sensor is responsible for the abnormal data values. To locate the correct sensor, a database of sensor metadata including the sensor’s serial number and other categories of supporting metadata can be developed (17). Sensor placement is critical; “incorrect sensor [light] placement can compromise system performance, cause discomfort to occupants and diminish savings” (18).

Kazado et al. makes the case that “the BIM model cannot show real-time information related to the performance of the building in the operational stage.” (19). Yet, an excellent example of BIM + BAS integration is Dasher 360 developed by Autodesk (20). Dasher 360 collects data from a wide range of sensors including temperature, RH, CO2 concentration, sound level and light intensity sensors. However, the loss of a large amount of data values one month should have been announced by warnings, either somewhere on the main interface or in the sensor list, in order to figure out the cause of this data value loss. Sensors have also been put in social housing and comfort parameter readings taken within the building information model (21), and augmented reality used to visualize sensor readings by programming Dynamo in Revit and using Unreal Engine (22). Another project used light sensors to control shades and louvers in BIM through Dynamo rather than the Revit API (23). BIM + sensor data is an expanding research area; these are just a few examples.

Fault detection has been successful in many engineering domains including automotive and industrial manufacturing for decades, but its application in AEC industry is still under development (24). Real-time data collection for detecting abnormalities is important ensure the proper function of facilities and equipment. Fault detection is achieved through algorithms. Some of them are quite simple and straightforward, and many are advanced types such as the statistical generation model (SGM) (25), adaptive-neuro fuzzy inference system (ANFIS) (26), and machine learning algorithms, for example, a “prediction-correction” process (24).

A fault detection algorithm does not have to be complex to be useful. It just has to be accurate within the bounds of its context. For example, a null reading from a temperature sensor in a room is also a fault if a value is expected. Determining the reason for the null reading might be complex, but the algorithm (check for null readings) is not. Another example is checking for a value within a specified range. If the value is not in the range, it is a fault. For example, a temperature set point might be 72 oF (22.2 oC) and the float 3 oF (20.6 oC to 23.9 oC) to achieve comfortable working conditions. A temperature reading of 80 oF (26.7 oC) would be considered a fault in this case, but perhaps not for another set of conditions/ranges.

The code that supports the AIOR was developed in Revit Macro IDE. An application level macro is desirable to ensure that the tool can be opened in more than one Revit session and all necessary references have been loaded automatically when a new macro is created.

Two types of connections can be found in the development process. A unidirectional connection exists between beacon sensors (hardware) and Adafruit IO Server (cloud-based platform). Sensors collect data and upload them to the server. In contrast, the interaction between the cloud-based platform and software is bidirectional. AIOR initiates requests for synchronization to retrieve data from the server, and concurrently, the server provides data files by downloading data to a local folder and importing to AIOR. Not all data from the server is necessary. For example, these IoT sensors used in the research collect real-time temperature, relative humidity (RH), and voltage data. Only temperature and RH data were used. These values are combined with pre-defined parameters including setpoints and comfort zone ranges (threshold values) to generate alerts when there are sensors that fault. There were nine main features coded, divided into three categories: file and user settings, data display and visualization, and comfort zone (Figure 2).

The tool development was broken into three sections: data retrieval and pruning, fault definition, and sensor model visualization (Figure 2). Features in AIOR were realized either using information from only one section (e.g. feature #1 - #4, #6 - #9), or combining information from multiple sections (e.g. feature #5).

Watt Hall on the USC campus was selected as the case study building. Five IoT sensors were installed in different locations for data collection. These sensors were originally installed for a project named TrojanSense (28). Temperature and RH data in Watt Hall third floor southeast corner (building science corner), upper Rosendin Gallery (two sensors), Watt 212, and Watt B1 was gathered and at the same time uploaded to Adafruit IO server. The intent was to demonstrate that the data from the five sensors could be retrieved by the Adafruit IO Reader, values shown in Revit as both tabular and line plots, and faults detected and highlighted in color with the corresponding sensor and its location.

The hardware used in the case study were developed by the TrojanSense team. They consist of ASAIR AM2302 IoT sensors powered by AA batteries and connected to Arduino ESP32 board with a Wi-Fi and Bluetooth microcontroller. These sensors collect temperature, relative humidity, and voltage data. Sensors were attached to the outside of the 3D-printed shells. Batteries supplied power to ESP32 boards and microcontrollers. These parts were put into the shells and installed under a studio desk, on a fire alerting device in upper Rosendin, on the staircase connecting lower and upper Rosendin, under the desk in Watt Hall 212, and on the wall in Watt Hall B1.

The Revit model was provided by USC FMS. The opacity of some building elements and the viewing angles are adjusted to ensure that sensor models are visible. The model of IoT sensors were created as a Revit family to be added to the model based on “Sensor Photos” and measurement of sensor dimensions.

To work with Adafruit IO Reader, the first step is to get program files and copy them to the specified folder. The program is started with Macro Manager in Revit, in the “Manage” ribbon. First, select “Application” tab control and any macro from A to E. Then, click on “Run” button to run the selected macro. Running the macro brings up the highlighted sensor model in a 3D view and the main interface of Adafruit IO Reader with real-time sensor data retrieved from the server.

In the main interface of the Adafruit IO Reader, most of the space on the main interface window is used for data table view panels and line plot view panels for visualizing sensor data. The interface is comprised of 12 modules: