Method for calculation of the risk of failure in a lighting system

Abstract

The present invention relates to a method to assist in the management and planning of maintenance operations of a lighting system, with particular reference to a street lighting system. The method provides the use of a predictive model for the calculation of an index representing the probability of the failure risk of a lighting system within a predetermined time interval, which does not require the analysis of data sensed at the devices constituting the lighting system.

Description

[0001] METHOD FOR CALCULATION OF THE RISK OF FAILURE

[0002] IN A LIGHTING SYSTEM

[0003] Cross-Reference to Related Applications

[0004] This Patent Application claims priority from Italian Patent Application No. 102023000013404 filed on June 28, 2023, the entire disclosure of which is incorporated herein by reference.

[0005] Technical Field

[0006] The present invention relates to a method to assist in the management and planning of maintenance operations of a lighting system, with particular reference to a public street lighting system. The method provides the use of a predictive model to calculate an index representing the probability of failure risk of a lighting system within a predetermined time interval (MTTF index, Mean Time To Failure). The index is calculated by means of geo-informatic analysis of variables connected to open or pre-existing databases characterising the territorial context wherein the system is located (exogenous variables) and the components and technologies the system itself is composed of (endogenous variables).

[0007] Background of the Invention

[0008] There are several known methods for the predictive management of maintenance operations of lighting systems, used to try to optimise the planning of these operations so as to reduce the downtime of the system, rather than to reduce the expenses associated with maintenance or repair activities. These methods are often based on the individual devices that make up the system (e.g. lighting body, support or switch board) and on the development and validation of models specifically built for that component through the processing of data sets relating to the structural / functional specifications of the same device / component, which comprise data sensed directly on the spot, by suitably prepared sensors. Such sensors are often included within the lighting devices, or are arranged within holders associated thereto.

[0009] The need to install sensors placed on board, rather than placed very close to lighting devices, causes an increase in the overall system costs, both for the installation as well as maintenance thereof. The increase is related to the higher cost of the devices and the higher cost of suitably qualified personnel to operate them. In addition, the presence of sensors results in an increase in the complexity of the system. Consequently, the pervasive monitoring of all the components necessary for the public lighting of entire cities is hardly feasible.

[0010] Summary of the Invention

[0011] The technical problem set forth and solved by the present invention is therefore to provide an improved method to assist in the prediction and planning of maintenance operations for a lighting system comprising a plurality of lighting devices, which allows to overcome the aforementioned drawbacks with reference to the prior art. The proposed method provides as output a representative index of the probability of failure risk of a lighting system within a predetermined time interval.

[0012] A definition of such a method is formulated in claim 1 , while preferred features of the invention are provided in the dependent claims.

[0013] The present invention achieves some significant advantages over the methods of the prior art. One of the main advantages is that the proposed method makes it possible to obtain an indication of the probability of the failure risk in the event of unavailability of data that can be sensed directly by, or at, the lighting system. The method, in fact, can take into analysis open databases for the characterisation of the territory, and / or pre-existing databases for the characterisation of the system (such as, for example, those referring to the census of public lighting, which provides information on the location of lighting points and the relative type of support and lamp). Therefore, there is no need to provide for dedicated sensors to be installed at, or integrated into, lighting devices. Of course, the method also allows to integrate additional data, where available, such as, for example, historical data on failures occurred in the lighting system.

[0014] Further advantages, features and methods of use of the present disclosure will become evident from the following detailed description of some embodiments thereof, given by way of non-limiting example.

[0015] Brief Description of the Drawings

[0016] Reference will be made to the attached Figures 1 to 4, which respectively show an example of a flowchart indicating the process of processing input data by a respective preferred embodiment of a model for calculating the MTTF index according to the present invention.

[0017] Various embodiments and variants of the invention will be described hereinafter, with reference to the flowcharts in the Figures introduced above.

[0018] Description of Embodiments of the Invention

[0019] The present invention provides a method for calculating an index representing the probability of failure risk of a lighting system within a predetermined time interval. This index is of the KPI (Key Performance Indicator) type, and consists of a numerical value that corresponds to a probability of the failure occurrence risk / need for maintenance in a predetermined time interval, wherein the association between the index value and the probability is coded according to predetermined rules.

[0020] The method is performed by a computer and involves first of all providing the computer with input data, which are data referring to the lighting system under investigation.

[0021] Input data can be fetched from databases external to the computer by wired or wireless communication means included in the computer, subject to the acquisition of any access authorisations, or can be entered manually by a user via an interface system connected to the computer itself. In addition, it is possible to envisage that the method for its execution provides the automatic acquisition of first input data from databases managed by third parties (e.g. databases of data relating to the weather conditions in the area where the system is located) and in combination requires the input of second input data by the user, who manually types or copies them into the interface system (e.g. data relating to the type of bulbs used in the system).

[0022] Data may comprise the type of lighting technology used and / or the age of the system, and / or historical data on the weather conditions at the installation site of the lighting system.

[0023] This data does not include data sensed by sensors - more generally, by any fit- for-purpose means - integrated, or installed on-board, or locally near one or more of the lighting devices included in the system. Therefore, it is neither provided nor necessary for the method to process data sensed on the spot to obtain the determination of the failure risk index.

[0024] The input data is processed by a statistical analysis model that provides as output an MTTF (Mean Time To Failure) index representing the mean time of failure occurrence in a lighting system device. Broadly speaking, failure means that a system lighting device breaks down or otherwise needs to be replaced or submitted to maintenance operations.

[0025] According to preferred variants, the input data include historical data of the failures that have occurred in the system and / or data related to the lighting system. In particular, input data may include one or more of the following:

[0026] - historical data relating to the meteorological conditions of the installation area of the lighting system;

[0027] - historical data relating to the risk of vandalism in the area where the lighting system is installed;

[0028] - topography of the territory;

[0029] - position of the street lamps;

[0030] - dimensional characteristics and configuration of lighting devices;

[0031] - type of bulb included in the lighting devices; technical sheets of the bulbs; macro-features of the electrical network serving the lighting system;

[0032] - level of reliability of the maintenance organization;

[0033] - level and age of historical maintenance work.

[0034] An input data processing model can be, for example, the Bayesian model, which uses input data to define the conditional probability of failure of street lighting and the associated greatest impact causes. This model is configured to provide the fault-tolerance capacity of the lighting system from a macro-level perspective, i.e. not predicting them with reference to a single street lamp but to a group of elements, in terms of exposure to a predetermined set of failures (electrical failure, mechanical failure, vandalism, etc.).

[0035] Alternatively, a specific model, with reference to a specific target system, may be developed by exploiting the Machine Learning technique, whereby training data for the implementation of the model are datasets of the lighting system referring to a predefined time interval, also comprising the data of failures occurred in the same time interval.

[0036] The accuracy in predicting the failure risk achievable by models developed with Machine Learning techniques could be higher than that achievable with Bayesian models consistently with the depth of the historical dataset, since they would allow to obtain information on the system at the micro level, i.e. at the level of the individual lighting device (e.g. individual street lamp), in terms of exposure to a predetermined set of failures and the associated level of reliability of each result obtained. However, given the need for large volumes of historical data for the increase in accuracy that can be managed with machine learning models, a hybrid approach - wherein preliminary analysis can be addressed with statistical / probabilistic models - is appropriate. For instance, Bayesian processing models can be provided in the first instance, subsequently “reinforced” with Machine Learning models, where an extensive history is available. In other words, preferred variants of the proposed method provide, in combination, the application of a statistical analysis model of the input data, e.g. Bayesian, and the successive application of a data processing model using Machine Learning techniques. According to preferred variants, Figures 1 , 2, 3, 4 show flow diagrams indicating the input data processing processes that may be employed by respective preferred embodiments of models for the calculation of the MTTF index according to the present invention.

[0037] According to further variants, for the determination of the input data processing model and calculation of the MTTF index, a machine-learning classifier trained to perform a statistical processing of the input data may be used, whereby the model is constructed by means of one or more computers by processing a plurality of training data sets, wherein each set comprises data of the same type as the input data already mentioned, relating to at least one lighting system and associated with historical data of failures occurred in the same lighting system. The machine learning classifier may comprise a neural network.

[0038] In any case, for processing the model, the method may comprise a step of collecting training data and pre-processing the data, which provides one or more of the following activities:

[0039] - normalisation of data (e.g. determination of thresholds, benchmarks, etc.);

[0040] - categorization according to predetermined aggregation criteria, and for each input data, association of a variable value weight, depending on the category in which the data is aggregated.

[0041] The step of pre-processing data can exploit processes of the extract, transform and load (ETL) type, which form combinations of data from multiple sources in a large central repository also referred to as a data warehouse. ETL processes use a set of business rules to clean and organise raw data and prepare it for storage, analysis and machine learning (ML).

[0042] Following the step of pre-processing the training data, the following model definition steps are provided:

[0043] - determination of characteristics and output provided by the model;

[0044] - sensitivity analysis to stress the model;

[0045] - testing the model output against a “real case” to check its reliability. According to preferred variants of the invention, the model can exploit a flow diagram for evaluating the Probability Mass Functions (PMFs) for the main elements of the devices constituting the lighting system (e.g. lamp, holder and switch board) as a function of input data, for the calculation of the final conditional probabilities. Thus, the evaluation of the corresponding Probability Density Functions (PDFs) approximating the PMFs is carried out, followed by the calculation of the Mean Time To Failure (MTTF) and failure probabilities. In the context of probability theory, the probability mass function or discrete density of a discrete random variable X, also encoded as p(X=x), is a real variable function that assigns to each possible value of X the probability of the elementary event.

[0046] Following the step of pre-processing the input data, the method may provide output generation and representation steps. A first step may be the definition and calculation of KPI indexes (Key Performance Indicators). These KPIs can be evaluated in relation to the user experience of the tool. When determining the model, in order to verify the results obtained in relation to the calculation of the KPIs, these indexes can be calculated for a set of test cases and compared with real event values.

[0047] The present invention has been described so far with reference to preferred embodiments. It is to be understood that there may be other embodiments that relate to the same inventive nucleus, all falling within the scope of protection of the claims provided below.

Claims

CLAIMS1. A method for predictive calculation of an index (MTTF) which represents the probability of failure risk in a lighting system within a predetermined time interval, wherein said lighting system comprises a plurality of lighting devices, said method being performed using a computer and comprising the following steps: provide the computer with input data including taxonomy data of the lighting system, including the type of lighting technology and / or the age of the system, and / or data relating to the weather conditions of the installation site of the lighting system, wherein said input data is fetched from databases external to the computer by wired or wireless communication means included in the computer, or manually entered by a user into an interface system connected to the computer; apply a statistical analysis model of said input data which provides as output the index (MTTF) which represents the probability of failure risk of a lighting device of the lighting system within a predetermined time interval, wherein said input data does not include data sensed by, or at, said lighting devices.

2. Method according to the preceding claim, wherein said input data includes historical data of the failures occurred in the lighting system.

3. Method according to claim 1 or 2, wherein said model is implemented by means of a machine learning classifier trained by one or more computers by processing a plurality of data sets, wherein each set comprises said input data relating to at least one lighting system associated with historical data of failures that occurred in the same at least one lighting system.

4. Method according to any of the preceding claims, wherein said input data includes one or more of the following:- historical data relating to the meteorological conditions of the installation area of the lighting system;- historical data relating to the risk of vandalism in the area where the lighting system is installed;- topography of the territory;- position of the street lamps;- dimensional characteristics and configuration of lighting devices;- type of bulb included in the lighting devices; - technical sheets of the bulbs;- level of reliability of the maintenance organization;- level and age of historical maintenance work.

5. Method according to claim 3 or 4, wherein said machine learning classifier comprises a neural network.

6. Method according to any of the preceding claims, comprising one or more of the following steps for pre-processing of input data:- normalization; categorization according to predetermined aggregation criteria, with association to each input data of a variable weight, depending on the category in which the data is aggregated.

Images