Air ventilation prediction system

Abstract

A computer-implemented method, system, and computer-readable medium, for providing guidance for mitigating spread of an aerosol and / or a gas, the method comprising receiving, via a user interface, environmental parameter values for an environment, providing the environmental parameter values to a machine-learning model trained using a plurality of evaluation scores, generating, by the machine-learning model and based on the environmental parameter values, guidance for mitigating the spread of an aerosol and / or a gas in the environment; and presenting the guidance to a user.

Description

[0001] Air Ventilation Prediction System

[0002] Field of the Invention

[0003] The invention relates generally to an air flow prediction system, and relates more specifically to generating guidance for mitigating the spread of an aerosol and / or gas in an environment.

[0004] Background

[0005] The Covid-19 pandemic posed a serious concern about the quality of the air around us, in terms of the transmission of airborne viral agents. In the early stages of the pandemic, the concern was more with larger size droplets from infected individuals and transmission via surfaces, but evidence implicating airborne (aerosol) transmission in the rapid spread of the disease emerged later, having shifted the focus to airborne transmission pathways. As a result, there are numerous studies investigating aerosol production and transport in various modes, and their infectivity. These include, but are not limited to, experimental measurement to characterise human-produced droplets during speaking and singing, computational analysis of aerosol spread in hospital spaces, survival of viral agents and infectivity in aerosol and in- situ monitoring of the actual viral transmissions. Among those, computational modelling of the flow that carries virus-laden aerosols caught huge attention worldwide because of the expectation that this type of analysis could help understand detailed associations between the aerodynamic circumstances, e.g. room size, ventilation, etc., and the aerosol dispersion hence the risk of disease transmission.

[0006] Summary

[0007] The invention is defined in the appended set of claims. Further example embodiments are set out in the detailed description.

[0008] In a first aspect, there is provided a computer-implemented method for providing guidance for mitigating spread of an aerosol and / or a gas, the method comprising: receiving, via a user interface, environmental parameter values for an environment; providing the environmental parameter values to a first machine-learning model trained using a plurality of evaluation scores, wherein, each evaluation score of the plurality of evaluation scores is associated with a simulated flow and / or aerosol density patterns of a plurality of simulated flow and / or aerosol density patterns, each simulated flow and / or aerosol density pattern of the plurality of simulated flow and / or aerosol density patterns is associated with a respective environment, and the first machine-learning model is configured to generate guidance for mitigating the spread of an aerosol and / or a gas based on the environmental parameter values; generating, by the first machine-learning model and based on the environmental parameter values, guidance for mitigating the spread of an aerosol and / or a gas in the environment; and presenting the guidance to a user.

[0009] Using a first machine-learning model to generate guidance for mitigating the spread of an aerosol and / or a gas in accordance with this method allows a user to be provided with optimal (retrofit) air ventilation solutions in real time, which are tailored to each user’s need, building conditions, and available resources. The use of a trained first machine-learning model eradicates the need for optimal solutions to be determined via a comparison of simulated flow and / or aerosol density patterns generated using computational fluid dynamics (CFD), or by experimental testing, both of which methods are time-consuming and resource-heavy.

[0010] In an example, the plurality of simulated flow and / or aerosol density patterns used to train the machine-learning model are generated by a second machine-learning model trained using another plurality of simulated flow and / or aerosol density patterns generated using a computational fluid dynamics (CFD) program, and the second machine-learning model is configured to generate a simulated flow and / or aerosol density pattern based on environmental parameter values.

[0011] By using a second machine-learning model to generate a plurality of simulated flow and / or aerosol density patterns, a large collection of ‘case studies’ can be rapidly generated for a plurality of environments. The time period needed to generate the plurality of case studies by the second machine-learning model is significantly shorter than would be needed if the same plurality of simulated flow and / or aerosol density patterns were generated using traditional CFD. Whilst generating a simulated flow and / or aerosol density pattern for a particular case (i.e., environment) using CFD may take between 1 and 2 days, the trained second machinelearning model may be able to generate the same simulated flow and / or aerosol density pattern within 1 to 2 seconds.

[0012] As the large collection of ‘case studies’ may be used to generate data for training the first machine-learning model, using a second machine-learning model to generate the case studies results in further reduction in the computational time and resources needed to provide optimal guidance for mitigating the spread of an aerosol and / or a gas in a new environment.

[0013] In another example embodiment, the aerosol comprises an airborne pathogen or contaminant. In such an embodiment, mitigating the spread of the aerosol reduces the risk of transmission of the pathogen or contaminant, and provides protection of persons in the environment against infection or otherwise unsafe exposure to the pathogen or contaminant. The aerosol may, for example, comprise Covid-19 pathogens. In another example, the aerosol may include particulate matter, such as dust or soot, which may cause damage to the eyes or lungs of a person present in the environment / room. The method of the claimed invention may be utilised in environments where pathogen transmission risk is high, such as, for example, hospital wards or consulting rooms.

[0014] In another example embodiment, each evaluation score of the plurality of evaluation scores indicates a risk to human health associated with the corresponding simulated flow and / or aerosol density pattern, and / or the respective environment. In such an embodiment, the guidance generated by the first machine-learning model sets out the optimal mitigation methods / systems for reducing a risk to human health posed by the aerosol or gas. The risk indicated by the evaluation score may be relative or actual, to enable easier comparison of the risk between environments, or provide further context, respectively.

[0015] In another example embodiment, the environmental parameter values define one or more of the following parameters: dimensions of a room; positions and / or dimensions of objects in the room; positions and / or number of doors in the room; positions and / or number of windows in the room; positions and / or number of air conditioning units in the room; positions and / or number of air cleaning units; and positions and / or number of people in the room. By providing the first machine-learning model with such parameters, which define the geometry and layout of a room for which mitigation guidance is required, the accuracy of the generated guidance per case may be improved.

[0016] In another example embodiment, the machine-learning models are deep-learning models. As deep-learning models are capable of identifying complex patterns, and can evaluate and refine their outputs for increased precision, they are advantageous in the present invention. The deep-learning models are able to generate outputs for increasingly complex environments (such as, for example, rooms with irregular geometries, with multiple people in, or rooms which are used for aerosol-generating purposes, such as endoscopy, ENT examinations, and dental treatment), and additionally can perform unsupervised learning during the training processes.

[0017] In another example embodiment, presenting the guidance to a user comprises providing the user with a ranked list of recommended mitigation methods. The mitigation methods may be ranked solely in terms of their effectiveness in preventing spread of the gas or aerosol in the room. Alternatively or additionally, other considerations may be taken into account when ranking the mitigation methods, such as the amount of power required for the mitigation method, or the cost associated with the mitigation method. As such, this embodiment enables a user to quickly identify the most suitable mitigation method from the guidance on a case-by- case basis, and taking into account various environmental and / or economic factors. In another example embodiment, presenting the guidance to a user comprises indicating an effectiveness of one or more mitigation methods. Quantifying the effectiveness of the generated guidance further enables a user of the system to make an informed decision regarding the mitigation methods / systems that can be put in place. In an example, whilst the generated guidance may indicate that one particular mitigation method would be the most effective in reducing transmission of an aerosol, a user of the system may wish to select a mitigation method that is indicated as being slightly less effective, as a trade-off between effectiveness and implementation cost.

[0018] In another example embodiment, the effectiveness of the one or more mitigation methods is based on a difference between the evaluation score associated with an environment defined by the environmental parameter values, and an evaluation score associated with the same environment additionally including the one or more mitigation methods. For example, the effectiveness may be provided a percentage decrease of an aerosol density when the one or more mitigation methods are used in the room. By indicating the effectiveness of each mitigation method in these relative terms, it may be easier for a user of the system to compare the effectiveness of different methods provided.

[0019] In another example embodiment, presenting the guidance to a user comprises generating an alert to indicate that a predicted aerosol concentration in the room exceeds a predetermined threshold value.

[0020] In this embodiment, a user of the system may be alerted in real-time that the aerosol concentration in the room is higher than recommended, such that immediate action to mitigate the aerosol concentration may be taken. Such an alert may be in the form of an audio alarm that sounds when an aerosol concentration exceeds a predetermined threshold value. Additionally or alternatively, a visual alarm in the form of a pop-up may appear on the user interface, for example warning the user that the room defined by the input parameters is at risk of having a high aerosol concentration. Such an embodiment may additionally be used in combination with one or more sensors for detecting aerosol concentration in the room. The one or more sensors may be provided, for example, to provide real-time feedback to enable auto-calibration of the mitigation guidance system.

[0021] In another example embodiment, the method further comprises receiving, via the user interface, an indication of available mitigation systems. By allowing a user to indicate that only certain mitigation systems are available, the number of outputs to be generated by the first machine-learning model, and output to the user interface, is significantly reduced, due to a reduction in the number of cases to be considered. For example, if the only mitigation system available is a small high-efficiency particulate absorbing filter, and this is indicated by the user to the first machine-learning model, then the first machine-learning model may only need to consider where best to position the filter in the environment / room when generating the guidance. This in turn results in a further reduction in the time and computing resources, such as processing power, required to generate the guidance that is provided to the user. Moreover the guidance output by the first machine-learning model may be directly applicable to real- world scenarios for which the environmental parameters are provided, including availability of certain mitigations.

[0022] In another example embodiment, the guidance comprises an indication of one or more mitigation systems. In another example embodiment, the guidance comprises suggestions for types of mitigation systems, placements for the one or more mitigation systems in the room, and / or an indication of the most effective mitigation system and placement. Rather than providing generic guidance on how to reduce the spread of an aerosol / gas in the room, the disclosed method is able to provide specific information regarding the exact systems / devices which will provide the greatest mitigating effects, and where they should be placed in the room to maximise their effectiveness. The precision and accuracy of the guidance provided is therefore greatly improved.

[0023] In another example embodiment, the one or more mitigation systems are retrofit mitigation systems. By providing guidance which implements retrofit mitigation systems (such as, for example, high-efficiency particulate absorbing filters, screens, or fans), the system and method provided becomes particularly suitable for use in existing buildings where significant structural works cannot be performed to employ built-in mitigation systems. In particular, the method may be particularly suitable for use in hospitals or other clinical environments, where operating rooms, consultation rooms and wards are in constant use, but where it is highly beneficial to mitigate the transmission of airborne pathogens.

[0024] In another example embodiment, the one or more mitigation systems comprises one or more of a high-efficiency particulate absorbing filter, an ultraviolet-C ‘IIV-C’ filter, a nonthermal plasma reactor, an air screen, a fan, air extraction device, air purifier or air-conditioning unit.. These filters (in particular high-efficiency particulate absorbing filters) are highly effective in mitigating the spread of aerosols and pathogens.

[0025] In another example embodiment, the method further comprises providing the guidance and the environmental parameter values to the second machine-learning model; and generating, by the second machine-learning model, one or more simulated flow and / or aerosol density patterns for the environment and implementing the guidance. In another example embodiment, the method further comprises presenting the one or more simulated flow and / or aerosol density patterns for the environment and implementing the guidance to a user. Generating a simulated flow and / or aerosol density pattern for the room in when the guidance is implemented, and presenting the simulated flow and / or aerosol density pattern to a user assists the user in understanding exactly where a recommend mitigation system should be placed in a room, in order to achieve optimum mitigation.

[0026] In a second aspect, there is provided a system for providing guidance for mitigating spread of an aerosol and / or a gas, the system comprising: a user interface; a memory; and one or more processors, configured to perform the method discussed above.

[0027] In a third aspect, there is provided a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the discussed above.

[0028] Brief Description of Drawings

[0029] One or more embodiments of the invention will now be described, by way of example only, and with reference to the following figures in which:

[0030] Figure 1 is a flow diagram depicting the claimed method.

[0031] Figure 2 is a flow diagram depicting further method steps.

[0032] Figure 3 is a diagram depicting the method steps involved in training first and second DL models.

[0033] Figure 4 is a depiction of an example user interface for use with the claimed method.

[0034] Figure 5 is a diagram of an example communication bus for a system suitable for executing the claimed method.

[0035] While the invention is susceptible to various modifications and alternative forms, specific embodiments are shown by way of example in the drawings and are herein described in detail. It should be understood however that the drawings and detailed description attached hereto are not intended to limit the invention to the particular form disclosed but rather the intention is to cover all modifications, equivalents and alternatives falling within the spirit and scope of the claimed invention. Any reference to prior art documents in this specification is not to be considered an admission that such prior art is widely known or forms part of the common general knowledge in the field. As used in this specification, the words “comprises”, “comprising”, and similar words, are not to be interpreted in an exclusive or exhaustive sense. In other words, they are intended to mean “including, but not limited to”. The invention is further described with reference to the following examples. It will be appreciated that the invention as claimed is not intended to be limited in any way by these examples. It will also be recognised that the invention covers not only individual embodiments but also combination of the embodiments described herein.

[0036] The various embodiments described herein are presented only to assist in understanding and teaching the claimed features. These embodiments are provided as a representative sample of embodiments only and are not exhaustive and / or exclusive. It is to be understood that advantages, embodiments, examples, functions, features, structures, and / or other aspects described herein are not to be considered limitations on the scope of the invention as defined by the claims or limitations on equivalents to the claims, and that other embodiments may be utilised and modifications may be made without departing from the spirit and scope of the claimed invention. Various embodiments of the invention may suitably comprise, consist of, or consist essentially of, appropriate combinations of the disclosed elements, components, features, parts, steps, means, etc., other than those specifically described herein. In addition, this disclosure may include other inventions not presently claimed, but which may be claimed in future.

[0037] It will be recognised that the features of the aspects of the invention(s) described herein can conveniently and interchangeably be used in any suitable combination.

[0038] Detailed Description

[0039] The spread of airborne aerosols or gases containing pollutants or contaminants is of great concern, not just in clinical environments such as hospitals, but in other environments where human exposure to a particular aerosol or gas is considered pertinent. Computational fluid dynamic (CFD) analysis of realistic conditions including aerosol dispersion, temperature effects, etc. in a large space such as a hospital wing is highly demanding in terms of computing power and time, meaning utilisation of CFD analysis for a specific environment is often infeasible. On the other hand, a supercomputing facility is not available for everyone, and also considering the recent advancement of machine learning techniques, prediction of flows using deep learning (DL) has come into various research areas including built environments. Such a model, often taking the form of physics-informed neural network (PINN), is expected to accelerate the process of flow modelling and prediction. Although these models have their own limitations such as requiring large volume of training data, once the training is completed, the runtime is short, for instance within 1 second to predict flow and associated physical variables.

[0040] According to the present disclosure, a computer-implemented method is provided which indicates optimal (retrofit) air ventilation solutions in real time, which are tailored to each user’s need, building conditions, and available resources. The use of a trained machine-learning model eradicates the need for optimal solutions to be determined via a comparison of simulated flow and / or aerosol density patterns generated using a CFD, or by experimental testing, both of which methods are time-consuming and resource-heavy.

[0041] Turning to Figure 1 , there is depicted is a flow diagram according to the present disclosure. At step 101 of Figure 1 , the method comprises receiving, via a user interface, environmental parameter values for an environment. The user interface may be implemented using a webbased application or via any other means such as on a local application. The environmental parameter values may be values representative of a room in a building, and define one or more of the following environmental parameters: dimensions of a room; positions and / or dimensions of objects in the room; positions and / or number of doors in the room; positions and / or number of windows in the room; positions and / or number of air conditioning units in the room; and positions and / or number of people in the room. The environmental parameters may include additional or alternative parameters, such as: mitigation systems / methods available to the user; whether or not a person in the room is infectious; temperature; humidity; amount of movement in the room (e.g., number of people walking around); strength and / or airflow speed of an air conditioning unit operation; breathing pattern of a person in the room (e.g., if they are coughing); and / or the general purpose of the room, including aerosol generating procedures performed such as endoscopy, ENT examinations, and / or dental treatment etc. The environmental parameter values may collectively define the environment.

[0042] It is noted that any of the environmental parameters provided in this disclosure may be used for training and as input for the first machine learning model and / or may be used for training and as input into the second machine learning model (such as that described with respect to figures 2 and 3).

[0043] At step 102, the method comprises providing the environmental parameter values to a first machine-learning model trained using a plurality of evaluation scores. Each evaluation score of the plurality of evaluation scores may be associated with a simulated flow and / or aerosol density pattern of a plurality of simulated flow and / or aerosol density patterns, and each simulated flow and / or aerosol density pattern of the plurality of simulated flow and / or aerosol density patterns may be associated with a respective environment. The evaluation score may be indicative of an aerosol density (e.g., a maximum aerosol density) in the associated flow and / or aerosol density pattern. In an example, the evaluation score may additionally or alternatively be indicative of a relative or actual risk to human health associated with the flow and / or aerosol density pattern, and / or the respective environment which is represented by the flow and / or aerosol density pattern. The machine learning model may, for example, be a deep- learning model however the method may alternatively be implemented using a different form of machine learning model.

[0044] At step 103, the method comprises generating, by the first machine-learning model and based on the environmental parameter values, guidance for mitigating the spread of an aerosol and / or a gas in the environment.

[0045] At step 104, the method comprises presenting the guidance to a user. In some examples, the guidance may be presented to the user via the user interface. However, the skilled person would recognise that other means for presenting the guidance to the user may be provided. An example user interface is provided below with respect to Figure 4.

[0046] In using a machine learning model trained using a plurality of evaluation scores to generate guidance for mitigating the spread of an aerosol and / or a gas in an environment, a user may be quickly provided with accurate guidance for mitigating the spread of an aerosol or gas in any environment, such that mitigations may be made in real-time. As such, the spread of any pathogens or pollutants to humans in the environment may be significantly reduced.

[0047] As mentioned above, the machine learning model which generates the guidance for mitigating the spread of an aerosol and / or a gas is trained using a plurality of evaluation scores. The evaluation scores could be generated in a number of ways, such as by analysing the results of a computational fluid dynamics (CFD) program. However in some cases, the machine learning model may be trained based on the output of another machine learning model. In particular, the plurality of simulated flow and / or aerosol density patterns associated with the plurality of evaluation scores used to train the first machine-learning model may be generated by a second machine-learning model trained using another plurality of simulated flow and / or aerosol density patterns generated using a CFD program, where the second machine-learning model is configured to generate a simulated flow and / or aerosol density pattern based on environmental parameter values.

[0048] Turning to Figure 2 there is provided a flow diagram depicting example further method steps, which may be used in conjunction with the method shown in Figure 1 . At step 201 , the method comprises providing the guidance and the environmental parameter values to the second machine-learning model. The other machine learning model having been trained to generate flow and / or aerosol density patterns using environmental parameter values and flow and / or aerosol density patterns generated using a computational fluid dynamics (CFD) program.

[0049] At step 202, the method comprises generating, by the second machine-learning model, one or more simulated flow and / or aerosol density patterns for the environment and implementing the guidance. At step 203, the method comprises presenting the one or more simulated flow and / or aerosol density patterns for the environment and implementing the guidance, to a user. In an example, the one or more simulated flow and / or aerosol density patterns may be presented to the user via the user interface. The skilled person would recognise that the one or more simulated flow and / or aerosol density patterns may be presented to the user by alternative means, such as, for example, a second user interface.

[0050] In utilising another machine learning model (itself trained on the results of a CFD program) to generate the aerosol distribution / flow patterns for presentation to the user, rather than using a CFD program, the patterns may be generated in a shorter space of time, thereby allowing for faster provision of useful guidance to the user.

[0051] Turning to Figure 3, there is provided a diagram depicting a method for training and utilising first and second DL models, in order to implement the method and system of the present disclosure. In a preferred embodiment, Model 1 in Figure 3 may be configured to implement the method steps performed by the second machine-learning model, as provided in Figure 2 and its associated description. Similarly, Model 2 in Figure 3 may be configured to implement the method steps performed by the first machine-learning model, as provided in Figure 1 and its associated description. While Figure 3 is described in relation to DL models, it should be appreciated that this approach is compatible with other machine learning models or more generic mathematical models. Furthermore, while Figure 3 describes various steps, it should be appreciated that these constitute merely one example and that these steps may be modified, and in some cases omitted, within the scope of the present disclosure. As such, the features of the methods shown in Figures 1 and 2 should be seen as compatible with the example of Figure 3, discussed below.

[0052] Step 1

[0053] In Step 1 , a plurality of three-dimensional (3D) aerosol distributions / flow patterns for a plurality of different environments are generated using computational fluid dynamics (CFD). The plurality of different environments are defined by a wide variety of inputs, which may, for example, fall under three categories of inputs: geometry; operating conditions; and mitigation. Inputs which fall within the ‘geometry’ category may, for example, include: room shape; room dimension; and room type. Inputs which fall within the ‘operating conditions’ category may, for example, include: number (an in some cases location) of people; and built-in ventilation. Inputs which fall within the ‘mitigations’ category may, for example, include: number and location of portable air purifiers and / or screens. The CFD simulations may be generated using any combination of inputs. A skilled person would recognise that additional and / or alternative inputs may also be used, such as the environmental parameters provided above, which may be used as inputs for the first machine-learning model of Figure 1 .

[0054] In an illustrative example, a room with particular width x height x length dimensions may be input as a base model into the CFD. The model may include one idealised human model, represented by a cylinder of a particular height and diameter, with its top rounded. The human model may have an opening at a particular elevation from the floor, which is a representation of the human mouth. The human model may breathe out a continuous or intermittent flow with a surrogate gas model of aerosol (e.g., 100% carbon monoxide (CO)). In an example, an intermittent flow may be used to imitate normal human breathing, speaking or coughing. The room may have an air conditioning flow inlet in the middle of a wall a particular distance from the ceiling, supplying fresh air, i.e. zero CO. The bottom end of the door may additionally be set as an opening as an outlet.

[0055] In the illustrative example, CFD simulations of aerosol dispersion in the model environment may be conducted using a CFD package, such as ANSYS Fluent (ANSYS Inc. Cannonsburg, USA). In this example, 3D incompressible Navier-Stokes equations may be solved in a discretised form in space and time using iterative numerical algorithms, with a turbulence model, such as using the SIMPLE algorithm for velocity-pressure coupling, 1st order implicit Euler scheme for time marching and standard k-e for the treatment of turbulence. Additional convection-diffusion equations may be solved for the aerosol phase, using CO as a surrogate. Density and viscosity of the air may be set to particular values, representative of the real-life values in the environment being modelled.

[0056] In the illustrative example, to use the aerosol mass concentration generated from CFD for the DL model for the deep learning model, the original CFD data - Cartesian (XYZ) coordinates and CO mass concentration in unstructured tetrahedral mesh (typically in the order of 10,000 to 1 ,000,000 nodes and elements) - may be resampled onto a 32x32x32 regular 3D grid. The aerosol concentration may be normalised by a set reference standard. Additionally, to allow the deep learning model to capture the geometry clearly, the geometrical data may be converted from the simple Cartesian coordinates (XYZ) to cylindrical coordinates (r-9z) taken from the center of the human model mouth. There are also 3 additional variables, namely the distances from wall (d_wall), the AC inlet (d_AC) and the door slit (d_exit).

[0057] A series of simulations with various human model locations and orientations, and exhalation velocities may be conducted in order to obtain a vast number of CFD simulations.

[0058] Step 2 Moving on to Step 2 of Figure 3, the plurality of inputs and corresponding 3D aerosol distributions / flow patterns may be stored in a first database, Database 1.

[0059] At least a subset of the plurality of inputs and corresponding 3D aerosol distributions / flow patterns may then be provided as training data to a machine-learning model, Model 1 , which may be a first deep-learning model. As above, Model 1 may be configured to perform the method steps associated with the second machine-learning model of Figure 2,

[0060] A second or remaining subset of the plurality of inputs and corresponding 3D aerosol distributions / flow patterns may be randomly selected and retained for model testing as unseen data. During model testing, the discrepancy between the CFD-generated and the machinelearning model-generated 3D aerosol distribution / flow patterns may be determined and analysed to evaluate the performance of Model 1.

[0061] The number of training samples provided to Model 1 may be continuously increased until a value of the discrepancy between the CFD-generated and the machine-learning modelgenerated 3D aerosol distribution / flow patterns plateaus.

[0062] In an illustrative example, the two following metrics may be used to evaluate the performance of Model 1 : Weighted Mean Squared Error (WMSE) and Normalized Logarithmic Mean Average Percentage Error (NLMAPE).

[0063] Here, y is the ground truth of the converted data, y is the model prediction, and n is total number of data points within the geometry. The WMSE may be the primarily metric in training. The NLMAPE may be used to facilitate a more intuitive understanding of the errors distribution, taking the normalized and logarithmic aerosol concentration distribution, such that the variations of concentration at a very low level can be depicted better.

[0064] Step 3

[0065] Moving on to Step 3 of Figure 3, Model 1 may be used to generate a plurality of 3D aerosol distribution / flow patterns using a plurality of inputs. The input parameters may be the same as those input into the CFD. Due to the previous training, Model 1 may be able to generate 3D aerosol distribution / flow patterns for unseen environments (defined by values for various environmental parameters), in which the input parameter values differ from the environments stored in Database 1 and used to train Model 1. The use of Model 1 to generate 3D aerosol distribution / flow patterns offers a fast prediction of aerosol in a short computational time and with limited computational resources. This allows rapid and comprehensive flow analysis including a large number of case studies. For example, in the scenario where prevention of airborne nosocomial disease prevention is explored, many of the scenarios including varied number of people present, built-in ventilation, and add-on mitigation such as portable air purifiers can be simulated within a short timeframe.

[0066] In utilising Model 1 (itself trained on the results of a CFD program) to generate the aerosol distribution / flow patterns which are later used to train Model 2 (see step 7 of Figure 3), a greater number of patterns may be generated in a short space of time, thereby allowing for faster gathering of training data for Model 2, and ultimately more reliable mitigation guidance.

[0067] The plurality of inputs and corresponding plurality of 3D aerosol distributions / flow patterns generated by Model 1 may then be stored in a second database. The plurality of 3D aerosol distributions / flow patterns generated by Model 1 include multiple flow patterns per single environment (e.g., per single room layout), each including a different mitigation system / method. The different mitigation systems / methods may include, as non-limiting examples, different ventilation systems, positions of ventilation systems, different particulate absorbing units and / or destruction units, sizes of particulate absorbing units and / or destruction units, positions of particulate absorbing units and / or destruction units, type of air supply diffusers, positions of air supply diffusers, positions of screens, sizes of screens, turning on an air-conditioning unit, turning off an air-conditioning unit, positions and directions of fans, opening a window, closing a window etc., and any combination of the above. Examples of particulate destruction units include units comprising ultraviolet-C (IIV-C) light and / or nonthermal plasma, which destroy or denature, rather than absorb pathogenic particles. A skilled person would understand that additional / alternative mitigation systems / methods may be provided as inputs.

[0068] Steps 4-6

[0069] Moving on to Steps 4-6 of Figure 3, each 3D aerosol distribution / flow pattern stored in the second database may be converted into / assigned an evaluation score. In a first example, the evaluation score may be determined using the maximum or mean aerosol density in a volume of the simulation space of the aerosol density pattern. This volume may be a spherical volume positioned around the head of a (non-infected) human model in the simulation space. In an example, the evaluation score may be a risk score, which indicates a relative or actual risk to human health associated with the corresponding aerosol distribution pattern. The risk score may be determined by identifying a quantitative aerosol density threshold, at or above which an infection / transmission risk is posed to a human in the simulation space. In such an example, the risk score may be a percentage of the room volume that is covered by the risky aerosol density level. A skilled person would recognise that additional and / or alternative methods for determining an evaluation score may be implemented.

[0070] Also assigned to each 3D aerosol distribution / flow pattern stored in the second database are further metrics associated with the corresponding mitigation method / system. These metrics may include, as a non-limiting example, an amount of electric power required by the mitigation method / system, and / or a cost associated with the mitigation method / system. The plurality of inputs, the evaluation scores associated with each 3D aerosol distribution / flow pattern and the further metrics may then be stored in a third database, Database 2. In an example, the 3D aerosol distributions / flow patterns may additionally be stored in Database 2.

[0071] Step 7

[0072] Moving on to Step 7 of Figure 3, at least a subset of the plurality of inputs, evaluation scores and further metrics may then be provided as training data to another machine-learning model, Model 2, which may be a second deep-learning model. As above, Model 2 may be configured to perform the method steps associated with the first machine-learning model of Figures 1 and 2,

[0073] In an example, a second or remaining subset of the plurality of inputs, evaluation scores and further metrics may be randomly selected and retained for model testing as unseen data. During model testing, the discrepancy between the evaluation scores stored in Database 2 and evaluation scores generated by Model 2 may be determined and analysed to evaluate the performance of Model 2.

[0074] The number of training samples provided to Model 2 may be continuously increased until a value of the discrepancy between the evaluation scores stored in Database 2 and evaluation scores generated by Model 2 plateaus.

[0075] Step 8

[0076] Moving on to Step 8 of Figure 3, Model 2, may be used to generate guidance regarding the best mitigation methods / systems for mitigating spread of an aerosol and / or a gas in an environment, based on a plurality of input parameters. The input parameters may fall under the categories of ‘geometry’ and ‘operating conditions’, and include the example parameters set out in the description of Step 1 of Figure 3, or in the description of Figure 4. Additionally, the input of Model 2 may also include an unranked list of available mitigation methods / systems.

[0077] Due to the previous training, Model 2 generates guidance regarding the best mitigation methods / systems for mitigating spread of an aerosol and / or a gas in an unseen environment defined by the input parameters. The use of Model 2 for generating guidance for mitigating spread of an aerosol and / or a gas provides a reduction in computational time and computational resources, and thus allows rapid and comprehensive analysis of optimal mitigation methods / systems for a particular environment.

[0078] Step 9

[0079] Finally, moving on to Step 9 of Figure 3, the inputs and outputs of Model 2 may be provided together to Model 1. Model 1 may then generate one or more simulated 3D aerosol distributions / flow patterns for the environment defined by the ‘geometry’ and ‘operating condition’ input parameters, when implementing the one or more mitigation methods / systems suggested in the guidance output by Model 2. The one or more simulated 3D aerosol distributions / flow patterns output by Model 1 may be representative of the best case scenarios for mitigating the spread of an aerosol and / or a gas in the environment defined by the input parameters of both Model 1 and Model 2.

[0080] The outputs of Model 1 and Model 2 may be output to a user via a user interface. This output allows the user to determine how best to mitigate the spread of an aerosol and / or a gas in the environment being analysed.

[0081] Turning to Figure 4, there is provided an example user interface 400 for use with the methods described herein, such as the example methods shown in Figures 1-3. As is shown in Figure 4, in an example of the present disclosure, the user interface 400 may include an input information element 401. The input information element 401 may include a plurality of input fields, in which a user can provide environmental parameter values for a plurality of parameters which define an environment. The (environmental) parameters may include, for example, room dimensions / size, a number of people in the room, such as a number of clinicians and / or patients, a number of doors in the room, a number of windows in the room, a number of pieces of large furniture in the room, a number of air conditioning outlets in the room. The input fields may allow the user to type their parameter values in, or allow the user to select a parameter value for a drop-down menu. Other methods of user input may alternatively or additionally be provided. In further examples not depicted in Figure 4, the parameters may additionally or alternatively include temperature, humidity, amount of movement in the room (e.g., number of people walking around), strength and / or airflow speed of an air conditioning unit operation, breathing pattern of a person in the room (e.g., if they are coughing), and the general purpose of the room, including aerosol generating procedures performed such as endoscopy, ENT examinations, and / or dental treatment etc. The parameters may further include any of the environmental parameters provided above as inputs for the first machine-learning model (Model 2), the second machine-learning model (Model 1) or the CFD discussed with respect to Figures 1-3.

[0082] The input information element 401 of the user interface 400 may additionally include an interactive element, which allows the user to interactively position objects (or people) corresponding to the input parameter values. In particular, the interactive element may allow users to drag objects and / or people identified by the input parameter values into their respective positions in the room, in order to create a 2D and / or 3D representation of the room for which the analysis is being run. The interactive element may additionally enable the user to indicate that one or more people in the room are infectious (i.e. that said one or more people are expected to produce an aerosol or gas including a pathogen), for example, by clicking on a virtual object representing the infectious person in a 2D or 3D representation of the room.

[0083] In an example of the disclosure, the user interface 400 may additionally include a 3D representation 402 of the environment for which the analysis is being run. This 3D representation 402 may be provided to assist a user of the system in determining the accuracy of the input parameter values and / or the representation of the room provided in the input information element 401 . Additionally or alternatively, the 3D representation 402 may provide a visual representation of a room in which the generated guidance is implemented, to assist the user in determining where to place the suggested mitigation system(s).

[0084] In an example of the disclosure, the user interface 400 may additionally include an output summary element 403, in which the generated guidance is presented to the user. As is depicted in Figure 4, the output summary element may include, for example, a ranked list of recommended mitigation methods / systems, and / or an indication of the effectiveness of one or more mitigation methods / systems. The effectiveness of the one or more mitigation systems may be indicated by based on a difference between an evaluation score associated with the room being analysed (i.e., the room defined by the input parameter values), and an evaluation score associated with the same room when the one or more mitigation systems are provided.

[0085] In an example, as depicted in Figure 4, the effectiveness may be indicated by a percentage, which is representative of a relative discrepancy between the evaluation score of the room being analysed, and the evaluation score with the same room when the one or more mitigation systems are provided. When a person in the room is infected with a virus, such as Covid-19, the percentage may represent a relative or actual reduction in transmission risk when the one or more mitigation methods / systems are implemented.

[0086] In another example, the effectiveness of the one or more mitigation systems may additionally or alternatively be indicated by a graph indicating aerosol build-up predictions. The graph may have aerosol concentration on the y-axis and time in hours on the x-axis. The lines provided on the graph may indicate the aerosol concentration build up in the room (or a particular location(s) within the room) when no mitigations are provided, and when one or more mitigation methods / sy stems are implemented.

[0087] In an example of the disclosure, the user interface 400 may additionally include an aerosol concentration map comparison element 404. As is depicted in Figure 4, the aerosol concentration map comparison element 404 may provide simulated aerosol density patterns for the room being analysed, in which the top two suggested mitigation systems are implemented. The simulated flow and / or aerosol density patterns may use a colour scale to demonstrate different aerosol densities, such that the user of the user interface can directly compare the density patterns, including the areas of highest aerosol density, and the effectiveness of the two compared mitigation methods. In an example, the aerosol concentration map comparison element 404 may use a red-blue colour scale to indicate aerosol densities, with red indicating a high aerosol density and blue indicating a low aerosol density.

[0088] The skilled person would understand that the elements of the user interface 400 may vary from those depicted in Figure 4 and discussed above. Any suitable means of depicting guidance for mitigating the spread of an aerosol and / or gas in an environment may be provided. Similarly, any suitable alternative for inputting environmental parameter data into a user interface computer system may be used with the claimed invention.

[0089] Figure 5 is a diagram of an example communication bus 500 for a system suitable for executing the methods disclosed herein, such as the methods of Figures 1 , 2, and / or 3. The communication bus 500 may include a memory 501 , one or more processors 502, a storage device 503, and one or more input / output (I / O) components 504 which may be used to provide a user interface, such as the user interface 400 shown in Figure 4. The communication bus 500 may additionally include further elements, not depicted in Figure 5.

[0090] The disclosed methods may be executed on one or more processors, wherein the one or more processors are comprised within a system or computer, alongside a user interface and a memory. The computer may include a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method as disclosed above. Instructions embedded or encoded in a computer-readable medium may cause a programmable processor, or other processor, to perform the method, e.g., when the instructions are executed. Computer-readable media may include non-transitory computer- readable storage media and transient communication media. Computer readable storage media, which is tangible and non-transitory, may include random access memory (RAM), read only memory (ROM), programmable read only memory (PROM), erasable programmable read only memory (EPROM), electronically erasable programmable read only memory (EEPROM), flash memory, a hard disk, a CD-ROM, a floppy disk, a cassette, magnetic media, optical media, or other computer-readable storage media. The term “computer-readable storage media” refers to physical storage media, and not signals, carrier waves, or other transient media. As noted above, computer readable media may include transient communication media. Such communication media may occur within a single computer system or between multiple computer systems, and may take the form of transient signal-conveying media such as carrier waves and transmission signals.

[0091] Therefore, from one perspective there has been described a computer-implemented method, system, and computer-readable medium, for providing guidance for mitigating spread of an aerosol and / or a gas, the method comprising receiving, via a user interface, environmental parameter values for an environment, providing the environmental parameter values to a machine-learning model trained using a plurality of evaluation scores, generating, by the machine-learning model and based on the environmental parameter values, guidance for mitigating the spread of an aerosol and / or a gas in the environment; and presenting the guidance to a user.

Claims

CLAIMS1. A computer-implemented method for providing guidance for mitigating spread of an aerosol and / or a gas, the method comprising: receiving, via a user interface, environmental parameter values for an environment; providing the environmental parameter values to a first machine-learning model trained using a plurality of evaluation scores, wherein, each evaluation score of the plurality of evaluation scores is associated with a simulated flow and / or aerosol density patterns of a plurality of simulated flow and / or aerosol density patterns, each simulated flow and / or aerosol density pattern of the plurality of simulated flow and / or aerosol density patterns is associated with a respective environment, and the first machine-learning model is configured to generate guidance for mitigating the spread of an aerosol and / or a gas based on the environmental parameter values; generating, by the first machine-learning model and based on the environmental parameter values, guidance for mitigating the spread of an aerosol and / or a gas in the environment; and presenting the guidance to a user.

2. The method of claim 1 , wherein, the plurality of simulated flow and / or aerosol density patterns associated with the plurality of evaluation scores used to train the first machine-learning model are generated by a second machine-learning model trained using another plurality of simulated flow and / or aerosol density patterns generated using a computational fluid dynamics (CFD) program, and the second machine-learning model is configured to generate a simulated flow and / or aerosol density pattern based on environmental parameter values.

3. The method of claim 1 or claim 2, wherein the aerosol comprises an airborne pathogen or contaminant.

4. The method of any preceding claim, wherein each evaluation score of the plurality of evaluation scores indicates a risk to human health associated with the corresponding simulated flow and / or aerosol density pattern, and / or the respective environment.

5. The method of any preceding claim, wherein the environmental parameter values define one or more of the following parameters: dimensions of a room; positions and / or dimensions of objects in the room; positions and / or number of doors in the room; positions and / or number of windows in the room; positions and / or number of air conditioning units in the room; positions and / or number of air cleaning units; and positions and / or number of people in the room.

6. The method of any preceding claim, wherein the machine-learning models are deeplearning models.

7. The method of any preceding claim, wherein presenting the guidance to a user comprises providing the user with a ranked list of recommended mitigation methods.

8. The method of any preceding claim, wherein presenting the guidance to a user comprises indicating an effectiveness of one or more mitigation methods.

9. The method of claim 8, wherein the effectiveness of the one or more mitigation methods is based on a difference between the evaluation score associated with an environment defined by the environmental parameter values, and an evaluation scoreassociated with the same environment additionally including the one or more mitigation methods.

10. The method of any preceding claim, wherein presenting the guidance to a user comprises generating an alert to indicate that a predicted aerosol concentration in the environment exceeds a predetermined threshold value.11 . The method of any preceding claim, further comprising receiving, via the user interface, an indication of available mitigation systems.

12. The method of any preceding claim, wherein the guidance comprises an indication of one or more mitigation systems.

13. The method of claim 12, wherein the guidance comprises types of mitigation systems and / or placements for the one or more mitigation systems in the environment.

14. The method of claim 12 or claim 13, wherein the one or more mitigation systems are retrofit mitigation systems.

15. The method of any of claims 12 to 14, wherein the one or more mitigation systems comprises one or more of a high-efficiency particulate absorbing filter, an ultraviolet-C ‘IIV-C’ filter, a nonthermal plasma reactor, an air screen, a fan, air extraction device, air purifier or air-conditioning unit.

16. The method of any of claims 2 to 15, further comprising: providing the guidance and the environmental parameter values to the second machine-learning model; and generating, by the second machine-learning model, one or more simulated flow and / or aerosol density patterns for the environment and implementing the guidance.

17. The method of claim 16, further comprising: presenting the one or more simulated flow and / or aerosol density patterns for the environment and implementing the guidance, to a user.

18. A system for providing guidance for mitigating spread of an aerosol and / or a gas, the system comprising: a user interface; a memory; and one or more processors, configured to perform the method of any preceding claim.

19. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any of claims 1 to 17.

Images