Lighting device operating state estimation device and method of estimating operating state of lighting device

Abstract

The present invention is directed to a lighting device operating state estimation device (100) for estimating an operating state of a target lighting device (502), configured to: provide respective materializations of a machine learning engine (104) with an initial set of parameters (P) to a plurality of lighting device providers (106, 108, 110); receiving engine data (Ea, Eb, Ec) from at least a subset of the plurality of lighting device providers, the engine data (Ea, Eb, Ec) indicating respective trained machine learning engines (104a, 104b, 104c) trained using respective proprietary training data sets (Ta, Tb, Tc) provided by the lighting device providers; a machine learning engine (105) that generates and provides updates using the corresponding engine data; and using the updated machine learning engine (105) and the current device data (D) to generate and provide device state data (M) indicative of an operating state of the target lighting device with increased reliability.

Description

Technical Field

[0001] The present invention is directed to a lighting device operating state estimation device, a lighting device operating state estimation arrangement, a method for estimating the operating state of a target lighting device, the use of a machine learning engine for executing the method for estimating the operating state of a lighting device, a training data set for training a machine learning engine to estimate the operating state, and a computer program. Background Art

[0002] Document CN110287640A describes a method and corresponding apparatus for estimating the lifespan of lighting equipment. The method includes obtaining a correlation model between multiple lighting devices, acquiring real-time data for a portion of the lighting devices, and identifying abnormal lighting devices based on the real-time data. The service lifespan of the abnormal lighting devices is determined, and based on the service lifespan, the lifespan of target devices associated with the abnormal lighting devices in the correlation model is estimated. Summary of the Invention

[0003] Reliability prediction of the remaining life of lighting assets (such as lighting fixtures) and reliability prediction of operational status to determine optimal maintenance strategies for lighting installations or lighting arrangements would be a key advantage for data-driven connected lighting solutions. However, in practice, many practical factors (available connectivity bandwidth, the cost of ingesting large amounts of sensor data in the cloud) as well as regulatory barriers (GDPR) hinder the sharing of collected data across different end users and countries. However, the quality of machine learning inferences depends on having a sufficient number of labeled examples for each failure mode. Because lighting fixtures are designed to be very reliable, they typically exhibit a failure rate of around 3% over a 10-year period. Therefore, labeled failures for training supervised machine learning models, such as in the case of CN110287640A, are difficult to obtain. However, each of the multiple maintenance modes of a lighting fixture requires a sufficient number of labeled cases (e.g., failures or maintenance issues) so that the machine learning model can reliably infer operational status or maintenance requirements when deployed in the field. In particular, short product lifecycles and constantly changing production and application conditions pose challenges to engineers responsible for developing reliable maintenance strategies for lighting fixtures in lighting installations or arrangements. Currently, lighting fixtures (including their LEDs and electronic components) are not kept in production for long periods of time (e.g., 10 years). Therefore, in order to perform machine learning, it is necessary to mine data on every possible maintenance issue or failure available. Similarly, today's lighting fixtures also use many newly developed 3D printing filament materials, which can be advantageously analyzed by machine learning techniques to identify mechanical or optical stability failure modes that are still unknown in the field at an early stage. Therefore, a cross-learning approach is needed across the expanded installed base of lighting fixtures.

[0004] It would be beneficial to improve the accuracy of estimates of the operating state of lighting devices.

[0005] According to a first aspect of the present invention, a lighting device operational state estimation device for estimating the operational state of a target lighting device is disclosed. The lighting device operational state estimation device includes a machine learning engine providing unit configured to provide corresponding specializations of a machine learning engine with an initial parameter set to a plurality of lighting device providers, wherein each lighting device provider provides corresponding lighting devices for one or more lighting facilities. Here, a specialization may refer to an instantiation or variant of an initial "base" model (i.e., a machine learning engine) with a specific parameter set (initial or updated parameter set). The lighting device operational state estimation device also includes an engine data receiving unit configured to receive corresponding engine data from at least a subset of the plurality of lighting device providers, the corresponding engine data indicating (i.e., originating from or based on) a corresponding trained machine learning engine, the trained machine learning engine having been trained using a corresponding proprietary training dataset, the proprietary training dataset including state data and / or operational data of a test lighting device provided by the corresponding lighting device provider. That is, the engine data (i.e., the updated parameter set) has been derived by training a machine learning engine with a corresponding proprietary training dataset, which includes state data and / or operational data of a test lighting device provided by the corresponding lighting device provider. The lighting device operational state estimation device also includes: a machine learning engine update unit, which is configured to generate and provide an updated machine learning engine using the corresponding received engine data; and an operational state estimation unit, which is configured to generate and provide device state data indicating the operational state of the target lighting device using the updated machine learning engine and current device data including the state data and / or operational data of the target lighting device. Thus, the device state data for a given target lighting device is estimated using an updated machine learning engine, which is generated using engine data provided by a subset of multiple lighting device providers and is not based solely on the corresponding lighting device provider's proprietary training dataset. This has the effect of increasing the reliability of the operational state estimate and, thus, enabling, for example, an optimized maintenance schedule.

[0006] The present invention relies on the understanding that machine learning techniques allow for the creation of detailed insights about failures, wear and product aging in order to flag quality issues in the field early and / or support future predictive maintenance use cases in lighting fixtures of a lighting installation, which can range from small smart home installations to large city-wide installations, such as those in the framework of Interact City or City Touch (e.g., lighting fixture installations with a large number of luminaires (such as the approximately 100,000 luminaires in Jakarta)). The concept of swarm-based machine learning (SBML) across independent lighting fixture providers opens up the opportunity to aggregate product reliability model learning across the entire lighting fixture range without introducing new privacy and security risks. Lighting fixture providers can include suppliers, operators of lighting fixtures, original equipment manufacturers, etc.

[0007] Therefore, the present invention relies on applying SBML to lighting devices (e.g., luminaires, controls, driver electronics, packaging, etc.) to provide a lighting device operational state estimation device based on a robust and continuously learning machine learning model. This estimation device can estimate the operational state and, based on this, recommend an optimal maintenance schedule. Specifically, the use of SBML technology allows for the integration of individual reliability model learning by participating lighting device providers (i.e., internal and external parties across many different markets, particularly original equipment manufacturers (OEMs), suppliers, nodes, etc.). The collaboratively trained SBML enables inference of faults in specific lighting devices installed in the field, thereby improving lighting system uptime and allowing for optimized maintenance planning.

[0008] The lighting device operational state estimation device is advantageously configured to provide a corresponding instantiation of a machine learning engine, including an initial parameter set, to each participating lighting device provider (e.g., an OEM) that provides corresponding lighting devices for one or more lighting installations. Thus, as an instantiation of the machine learning engine, the lighting device operational state estimation device provides a suite of machine learning analysis functions to perform analysis on a given dataset. Each lighting device provider trains its instantiation of the machine learning engine using a corresponding proprietary training dataset. Thus, the corresponding instantiation is trained using the training dataset, which includes state data and / or operational data of test lighting devices provided by the lighting device provider based on the specific usage conditions of the lighting devices. That is, each lighting device provider receives an initial "base model" or machine learning engine with an initial parameter set. This initial parameter set serves as a starting point for training updated instances / variants of their base model using the training dataset, which includes state data and / or operational data of test lighting devices provided by the lighting device provider based on the specific usage conditions of the lighting devices. The term "test lighting devices" refers to those lighting devices whose state data and / or operational data are used by the respective lighting device provider to train the corresponding machine learning engine. Thus, the test lighting fixtures associated with a given lighting fixture provider include at least a subset of the corresponding lighting fixtures provided by that provider. Sharing engine data obtained or generated by each of the lighting fixture providers broadens the types of luminaire failure modes learned by the updated machine learning engine, thereby enabling the updated machine learning engine to reliably infer the operational status of deployed lighting fixtures based on the current device data of the target lighting fixtures. The device status data generated and provided by the operational status estimation unit indicates the current operational status of the lighting fixtures and can indicate whether maintenance action is expected to be required for a given target lighting fixture within a given timeframe, thereby enabling better planning of maintenance service schedules. The device status data can, for example, relate to operating parameters of the lighting fixtures, such as drift in correlated color temperature or lighting intensity.

[0009] An updated machine learning engine is generated by merging the engine data provided by each of the lighting fixture providers to obtain a common engine. For example, the merging can be done by calculating the parameters as an average, a weighted average, or using a median algorithm of the parameters of the trained machine learning engines.

[0010] In a particular embodiment, the device status data is maintenance data indicating maintenance requirements for the target lighting device. Therefore, this particular embodiment is configured as a maintenance requirement estimation device for estimating maintenance requirements for the target lighting device, and includes a maintenance requirement estimation unit as an operating status estimation unit. The maintenance requirement estimation unit uses updated machine learning and current device data including status data and / or operating data for the target lighting device to generate and provide maintenance data indicating maintenance requirements for the target lighting device. The maintenance data may indicate a requirement for a maintenance action and include instructions for a service action (e.g., replacement, repair, cleaning, recalibration, relocation, etc.). Additionally or alternatively, it may include operational instructions for mitigating the cause or effect of the estimated fault, thereby extending the life of the lighting device and preventing complete interruption of operation of the lighting device before the service action occurs.

[0011] In a preferred embodiment, the lighting device operational state estimation device further includes a lighting control unit that is connected to the lighting device operational state estimation unit (or maintenance requirement estimation unit) or a portion thereof and is configured to use the device state data to provide operational instructions for operating the target lighting device. Thus, the knowledge (in the form of device state data) gained by applying current device data to the updated machine learning engine can be used to control the operation of the target lighting device, for example, operating it in a manner that can extend its expected lifespan (such as, for example, by limiting light intensity output, changing cooling parameters, preventing on / off cycling of the lighting device, etc.).

[0012] According to a second aspect of the present invention, a lighting device operating state estimation arrangement is disclosed, which includes a lighting device operating state estimation device according to the first aspect of the present invention and a plurality of machine learning engine units, wherein the plurality of machine learning engine units are configured to: receive a corresponding instantiation of a machine learning engine with an initial parameter set from the lighting device operating state estimation device; provide a corresponding proprietary training data set for training the machine learning engine, the corresponding proprietary training data set including state data and / or operating data of a test lighting device provided by a corresponding lighting device provider; determine an updated parameter set indicative of the trained machine learning engine using the proprietary training data set and the corresponding instantiation of the machine learning engine; and provide engine data indicative of the corresponding trained machine learning engine to the lighting device operating state estimation device.

[0013] The arrangement of the second aspect of the present invention may be implemented as a maintenance requirement estimation arrangement, wherein the device status data is maintenance data indicating maintenance requirements of the target lighting device.

[0014] Each of the machine learning units is associated with a corresponding lighting fixture provider and is configured to be trained using a corresponding proprietary training data set available to the corresponding lighting fixture provider.

[0015] Preferably, and to preserve sensitive information associated with proprietary training datasets, only updated parameter sets are provided to the lighting fixture operating state estimation device. While lighting fixture providers share their gradually refined updated parameter sets (e.g., the weights of the neural network that makes the operating state inference), these providers retain all their customer-, maintenance-, and product-specific information for themselves. Consequently, lighting fixture providers (e.g., OEMs) do not need to share any specific data and can benefit from improved estimates of operating state. This collaborative improvement of reliability models across multiple lighting fixture providers enables accelerated reliability learning, particularly across different global regions, product usage, and stress patterns, to which different lighting installations are subject. The SBML approach leverages machine learning to enable continuous adaptation of the reliability model.

[0016] The lighting device operating state estimation device and / or machine learning engine unit can be implemented as software in a corresponding computer system with output and input interfaces, which provide and receive data, namely the embodiment of the machine learning engine, engine data (such as updated parameter sets) and device state data.

[0017] In certain embodiments, the device status data is generated and provided by a machine learning unit associated with a plurality of lighting device providers.

[0018] Preferably, the device status data indicates maintenance actions expected to be required for a given lighting device. In one embodiment, the device status data additionally or alternatively includes operational instructions for operating the target lighting device to increase the expected operational life, for example, limiting the maximum light intensity, increasing cooling of the lighting unit, etc. The operational instructions can be provided to the corresponding lighting device provider or directly to the target lighting device to control its operation. In a preferred embodiment, when the device status data indicates the need for maintenance actions, the device status data includes an indication of the service action (e.g., replacement, repair, cleaning, recalibration, relocation, etc.) and operational instructions for mitigating the cause or effect of the estimated fault, thereby extending the life of the lighting device and preventing complete interruption of the operation of the lighting device before the service action occurs.

[0019] In one embodiment of the arrangement, the respective operating status estimation unit is comprised or owned by one or more lighting device providers, which can therefore use current device data of their lighting devices to obtain device status data or maintenance data.

[0020] A third aspect of the present invention is formed by a method for estimating the operating status and / or maintenance requirements of a target lighting device. The method comprises the following steps:

[0021] - providing respective instantiations of the machine learning engine with the initial parameter set to a plurality of lighting fixture providers, each lighting fixture provider providing respective lighting fixtures for one or more lighting facilities;

[0022] - Each lighting fixture provider provides a corresponding proprietary training dataset to the corresponding embodiment for training the machine learning engine, wherein the training dataset includes status data and / or operation data of the test lighting fixtures provided by the lighting fixture provider;

[0023] - Each lighting fixture provider uses the proprietary training dataset and the corresponding instantiation of the machine learning engine to determine an updated parameter set indicative of the trained machine learning engine;

[0024] - each lighting fixture provider provides engine data indicative of a corresponding trained machine learning engine for use in generating an updated machine learning engine; and

[0025] -Using an updated machine learning engine and current device data including status data and / or operating data of the target lighting device, generate and provide device status data (e.g., in the form of maintenance data) indicating the operating status or maintenance requirements of the target lighting device.

[0026] The method of the third aspect of the present invention shares the advantages of the lighting device operating state estimation device of the first aspect and the lighting device operating state estimation arrangement of the second aspect.

[0027] In the following, embodiments of the method of the third aspect will be described. These embodiments are suitable for being performed by corresponding embodiments of the lighting device operating state estimation device according to the first aspect of the present invention or corresponding embodiments of the lighting device operating state estimation arrangement according to the second aspect advantageously adapted thereto.

[0028] In one embodiment, blockchain is used to implement instances of the machine learning engine, trained machine learning engines, and updated machine learning engines. Due to the blockchain implementation, any data changes can be reliably retrieved, and no erroneous or malicious parameter sets can be inserted without traceability to the lighting fixture provider. Furthermore, future service contracts can be sold simply by generating new instances of the updated machine learning engine.

[0029] In one embodiment, the device status data generated additionally or alternatively includes operational lifetime data indicative of an expected remaining useful lifetime of the target lighting device.

[0030] Furthermore, preferably, and in order to ensure data privacy of the respective lighting device providers' proprietary training datasets, in embodiments of the method, the engine data provided by the lighting device provider includes, or preferably consists of, at least a portion of the updated parameter set. For example, a given lighting device provider may decide to share only a subset of the updated parameters (e.g., parameters related to the impact of lightning strikes on driver health, rather than parameters related to driver health degradation associated with cold starts of LED drivers).

[0031] Thus, lighting device providers (e.g., OEMs) only share the parameters of a trained machine learning engine or neural network, or a portion thereof, and not the training dataset itself. This approach requires significantly less communication bandwidth and reduces data ingestion costs for each lighting device provider's cloud. Furthermore, some lighting device providers may not accept their customer data being in the hands of another party. However, each participating lighting device provider desires to learn together with the other providers to improve estimates of the operating state or expected operating life of the target lighting device. Preferably, the updated parameter set is encrypted before being provided (particularly to the lighting device operating state estimation device) and shared with all other instances of the machine learning engine implemented in the remaining machine learning engine units associated with each of the remaining lighting device providers or OEMs participating in the collaborative, jointly owned SBML model. In this way, all partner engines maintain the same learning state without exchanging the training data itself.

[0032] In another embodiment, the method further comprises providing the updated machine learning engine to one or more of the plurality of lighting device providers and / or one or more third-party lighting device providers. Thus, for example, a service contract may be offered or sold to a non-participating lighting device provider or other third party for generating another instantiation of the updated machine learning engine. In one embodiment, the method comprises generating a corresponding updated machine learning engine and transmitting the corresponding updated machine learning engine to the participating providers or third parties. For example, one provider may choose not to participate in a certain aspect (such as useful life inference after a lightning strike) because they believe they already have sufficient expertise in that particular area. A second provider may participate and share information regarding this aspect. Thus, the updated machine learning engine provided to each of the providers may be different.

[0033] Preferably, the method includes providing a service to associated partners, and optionally also to end customers, in which information about the expected operating state or expected operating life is available. For example, a certificate (e.g. in the form of a QR code) detailing the inferences of the reliability model and / or the expected operating state or expected operating life can be provided. These certificates are useful for the second-hand market. The useful life predictions made by the group model need to have the following proof: they are both credible and provide confidence in the inferences. Due to the proposed SBLM-based collaboration across manufacturers or other lighting device providers, the SLBM-based model can learn from more labeled failures, resulting in a higher confidence in the remaining useful life inference than a single OEM can achieve alone. Since lighting devices only rarely fail, having observations about luminaire failures is very valuable.

[0034] In a preferred embodiment, the proprietary training dataset includes status data and / or operational data indicating one or more of lighting device metadata, environmental data associated with environmental conditions at an operating location of the lighting device, fault data indicating an operational fault of the lighting device, and service data indicating a service action performed on the lighting device, and may include, for example, a replacement action of any component of the lighting device, a repair action of any component of the lighting device, a cleaning action of any component of the lighting device, a recalibration action of any component of the lighting device, a change in the location of the lighting device, etc. In particular, the lighting device metadata indicates one or more of the type of the lighting device, the serial number of the lighting device, the production date of the lighting device, the production location of the lighting device, the operating or installation location of the lighting device, the usage of the lighting device (e.g., indoor, outdoor), the lighting device components of the lighting device (e.g., hardware components), and / or the driver type of the lighting device (e.g., software components).

[0035] Additionally or alternatively, in certain embodiments, the environmental data indicates weather conditions during operation of the lighting device at the installation location, and / or the environmental data indicates conditions to which the lighting device was exposed prior to installation (such as storage conditions). For example, the environmental data includes time-series weather data, including, for example, temperature data, humidity data, solar radiation data, wind speed data, icing data, and / or lightning strike data. The environmental data can be determined by sensors integrated into the respective lighting device or from dedicated external sensors. Storage conditions are also relevant, as lighting devices stored in high-temperature conditions may be damaged or have a shortened lifespan. Similarly, low-temperature exposure during storage of LED drivers may have caused components (such as SMD diodes and capacitors) to be pulled out of the PCB due to crystallization of the potting bitumen at low temperatures.

[0036] In another embodiment, the proprietary training dataset includes fault or performance data, including, for example, data indicating lux measurements, color, point in field, lifetime data, etc. The proprietary training dataset may also include image data or light sensor data indicating lux levels of one or more lighting fixtures, such as daytime and / or nighttime satellite images of streetlights that can be used to infer the current performance status of a given lighting fixture and the physical environment of each of the lighting fixtures within the image.

[0037] In another embodiment, the method further comprises the step of encrypting engine data indicative of a corresponding trained machine learning engine before providing the engine data for generating an updated machine learning engine (in particular, providing encrypted updated engine parameters as encrypted engine data).

[0038] Preferably, and to ensure accountability and traceability of device status data and / or maintenance data, the method may further include storing the device status data, particularly in a memory unit of the lighting device, particularly using blockchain technology. The device status data, which may also include the expected remaining lifespan of the lighting device, can be advantageously used to price the lighting device in the secondary market.

[0039] A fourth aspect of the invention is formed by the use of a machine learning engine or the lighting device operating state estimation arrangement of the second aspect for performing the method of estimating the operating state of a lighting device according to the third aspect of the invention.

[0040] A fifth aspect of the present invention comprises a training data set for training a machine learning engine, wherein the machine learning engine is configured to estimate an operating state according to the method of the third aspect. The training data set comprises state and / or operating data of a test lighting device, in particular state and / or operating data indicating one or more of lighting device metadata, environmental data associated with environmental conditions at an operating location of the lighting device, fault data indicating an operational fault of the lighting device, and service data indicating a service action performed on the lighting device.

[0041] According to a sixth aspect, a computer program is disclosed. The computer program comprises instructions which, when executed by a computing system of the lighting device operating state estimation arrangement according to the second aspect, cause the computing system to perform the method of the third aspect.

[0042] The present invention allows, among other things, enhanced predictions of potential failures, thereby improving maintenance planning for (street) lighting installations. Furthermore, the updated machine learning engine provides a solid foundation for estimating the remaining useful lifetime of lighting fixtures, thereby boosting the second-hand market for lighting fixtures (including similar drivers, LED engines, and luminaires). Collaboratively improved reliability models based on the updated machine learning engine (which leverages data provided by multiple lighting fixture providers) will accelerate product reliability learning across different world regions, product usage, and stress patterns (e.g., ambient temperature). The SBML approach leads to continuous adaptation of the reliability model via machine learning; thanks to the possibility of frequent retraining (e.g., the updated machine learning engine can be provided again as a corresponding instantiation of the machine learning engine to all participating providers for retraining using proprietary training datasets), the model can adapt to shifts in product usage over time, as well as drifts in production conditions and used materials. Furthermore, providers do not need to share training datasets and still benefit from improved reliability inferences for their installed lighting fixtures.

[0043] The present invention can be advantageously used to train machine learning engines to predict the remaining life of lighting assets and determine the optimal maintenance strategy for lighting facilities. It allows the use of labeled data (e.g., data collected from lighting fixtures and facilities across different end users and regions or countries) without the need to share data between independent enterprises.

[0044] It should be understood that the lighting device operating state estimation device of claim 1, the lighting device operating state estimation arrangement of claim 3, the method for estimating the operating state of a target lighting device of claim 4, the use of the lighting device operating state estimation arrangement of claim 13, the training data set of claim 14, and the computer program of claim 15 have similar and / or identical preferred embodiments, in particular as defined in the dependent claims.

[0045] It shall be understood that a preferred embodiment of the invention can also be any combination of the dependent claims or the above-described embodiments with the respective independent claim.

[0046] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In the following drawings:

[0048] Figure 1 A schematic block diagram of a lighting device operating state estimation arrangement according to an embodiment of the present invention and comprising a lighting device operating state estimation device and a plurality of machine learning engine units is shown;

[0049] Figure 2A flowchart showing exemplary training steps of a population-based machine learning (SBLM) implemented by the lighting device operating state estimation arrangement according to the present invention; and

[0050] Figure 3 A flowchart of a method for estimating an operating state of a target lighting device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0051] Figure 1 A schematic block diagram shows a lighting device operating state estimation arrangement 200 according to an embodiment of the present invention, comprising a lighting device operating state estimation apparatus 100 and a plurality of machine learning engine units 202, 204, and 206. Each machine learning unit is associated with a respective lighting device provider 106, 108, and 110. Each lighting device provider provides a respective lighting device 500a, 500b, and 500c, for example, for one or more lighting fixtures. Each of the plurality of machine learning engine units 202, 204, and 206 is configured to receive a respective instantiation of the machine learning engine 104 with an initial parameter set P from the lighting device operating state estimation apparatus 100. The instantiation 104 is provided via the machine learning engine providing unit 102. Each of the machine learning engine units is further configured to provide a corresponding proprietary training dataset Ta, Tb, Tc for training the machine learning engine 104. The proprietary training dataset Ta, Tb, Tc includes state data and / or operational data of the test lighting devices 500a, 500b, 500c provided by the corresponding lighting device provider 106, 108, 110. The proprietary training dataset includes state data and / or operational data indicating one or more of lighting device metadata, environmental data associated with environmental conditions at the operating location of the lighting device, fault data indicating operational faults of the lighting device, and service data indicating service actions performed on the lighting device. For example, the lighting device metadata may indicate one or more of the type of lighting device, the serial number of the lighting device, the manufacturing date of the lighting device, the manufacturing location of the lighting device, the operating location of the lighting device, the usage of the lighting device, the lighting device components of the lighting device, and the driver type of the lighting device. The environmental data may further indicate weather conditions during operation of the lighting device at the installation location or indicate storage conditions to which the lighting device was exposed prior to installation. The proprietary training dataset may also include image data or light sensor data indicating lux levels of one or more lighting devices. The proprietary training data may be determined or ascertained in part or in whole by a sensing device integrated into or associated with the lighting fixture. Additionally or alternatively, the proprietary training data or a portion thereof may be provided by an external sensor or system (such as, but not limited to, a weather station or satellite).

[0052] The machine learning units 202, 204, and 206 are further configured to use the corresponding proprietary training data sets Ta, Tb, and Tc and the corresponding instantiation of the machine learning engine 104 to determine the corresponding updated parameter sets Pa, Pb, and Pc indicative of the trained machine learning engine, and to provide the corresponding engine data Ea, Eb, and Ec indicative of the corresponding trained machine learning engine 104a, 104b, and 104c to the lighting device operating state estimation device. Therefore, even though each of the machine learning units 202, 204, and 206 is started with an instantiation or instance of the same machine learning engine having the same startup parameters (e.g., coefficients), they are trained using different proprietary training data sets Ta, Tb, and Tc that indicate the operation or state of the lamps managed by the corresponding lighting device provider. As a result, each instantiation of the machine learning engine evolves toward a different updated parameter set Pa, Pb, and Pc. For example, lighting device 500a provided by provider 202 is typically used in a humid tropical environment in an outdoor location, and the trained machine learning engine will tend to estimate failure modes or maintenance actions related to humid conditions with increased reliability; while lighting device 500b provided by provider 204 is typically used in an indoor location in Tucson, where conditions are much drier, and the possible failure mode is not significantly related to high humidity conditions. However, the engine data Ea, Eb, and Ec indicating the corresponding trained machine learning engine are provided back to the lighting device operating state estimation device 100. Engine data is received at an engine data receiving unit 112, which is configured to receive engine data Ea, Eb, Ec from at least a subset of the plurality of lighting device providers 202, 204, 206, the engine data Ea, Eb, Ec indicating respective trained machine learning engines 104a, 104b, 104c trained using respective proprietary training data sets Ta, Tb, Tc, the proprietary training data sets Ta, Tb, Tc including status data and / or operational data of the test lighting devices 500a, 500b, 500c provided by the respective lighting device providers.

[0053] The machine learning engine update unit 114 is configured to use the corresponding received engine data Ea, Eb, Ec to generate and provide an updated machine learning engine 105, which thus includes knowledge from each of the trained machine learning engines of the participating lighting device providers, thereby increasing the number of possible failure modes or maintenance requirements that the updated engine can identify. The operating state estimation unit 116 included in the lighting device operating state estimation device 100 is configured to use the updated machine learning engine 105 and the current device data D including state data and / or operating data of the target lighting device 502 to generate and provide device state data, such as maintenance data M indicating the operating state (e.g., maintenance requirement) of the target lighting device 502. The target lighting device can belong to any of the sets of lighting devices provided by any of the lighting device providers 106, 108, 110, or can belong to a non-participating entity (i.e., another provider that is not involved in the training and upgrading of the machine learning engine 104 and merely benefits from the knowledge provided by the expanded effective combined training data set).

[0054] Additionally or alternatively, the lighting fixture provider may include a corresponding operating status estimation unit for estimating operating status (e.g., maintenance requirements). Preferably, the updated machine learning engine 105 is also provided to one or more of the plurality of lighting fixture providers and / or to one or more third-party lighting fixture providers. The updated machine learning engine is then run on each of the machine learning engine units associated with each of the lighting fixture providers, which can then benefit from the expanded situation database (e.g., based on data obtained from all sets of lighting fixtures 500a, 500b, 500c) that has resulted in the updated machine learning engine 105.

[0055] Therefore, using the updated machine learning engine 105 at the lighting device operating state estimation device 100 or at any one of the machine learning engine units 202, 204, 206, together with using the current device data D of any given target lighting device 502 (which may or may not belong to any lighting device provider 106, 108, 110), the operating state of the target device 502 can be estimated with increased reliability because the updated machine learning engine 105 is generated from a larger database than any one of the trained machine learning engines 104a, 104b, 104c of a separate lighting device provider.

[0056] Preferably, and in order to reduce the bandwidth necessary for the arrangement 200 and ensure the confidentiality of the proprietary training data sets Ta, Tb, Tc, the engine data Ea, Eb, Ec provided by the lighting device provider consists only of the corresponding updated parameter sets Pa, Pb, Pc or at least a subset thereof.

[0057] Preferably, the device status data M includes operational lifetime data indicating the expected remaining useful lifetime of the target lighting device, and even more preferably, the device status data is stored on a memory unit of the lighting device of the lighting device provider and / or on a database associated with the lighting device provider, in particular using blockchain technology.

[0058] In one exemplary arrangement 200, an updated machine learning engine is used to infer the remaining useful life of a specific target lighting fixture that has been installed in the field. In this particular example, all lighting fixtures provided by a provider utilize the same driver hardware and the same driver firmware, while other aspects, such as the optics of the lighting fixtures, vary between providers.

[0059] In another exemplary arrangement 200, an updated machine learning engine is advantageously used to infer—e.g., on a fleet basis—the average remaining useful life of an outdoor streetlight installation (e.g., a City Touch installation) comprising a large number of luminaires (e.g., approximately 100,000 luminaires in Jakarta). Information about the remaining useful life can be used, for example, to price contract renewals for the lighting installations. For example, using the updated machine learning engine, it can be inferred that between years 10 and 15, 25% of the luminaires in the installed lighting installation under the analyzed project will experience a failure or operating condition and / or maintenance requirement, and this information can be used to adjust the price.

[0060] In a further development of arrangement 200, a newer machine learning engine is used to infer the remaining useful life of a specific lighting fixture. This information can be stored on the luminaire, for example, in a blockchain, to ensure accountability and traceability. The remaining useful life information is then used to price the luminaire in the second-hand market.

[0061] In a further development of arrangement 200, an updated machine learning engine is advantageously used to infer the Level 1, Level 2, and Level 3 carbon footprints of different operation and maintenance scenarios of lighting fixtures, particularly streetlights. For example, a motion sensor may frequently switch the luminaire, which saves energy and thus reduces Level 1 carbon emissions, but the frequent switching results in a shortened remaining useful life, thus leading to increased Level 3 carbon emissions.

[0062] In a further development of this arrangement, lighting devices 500a, 500b, 500c include at least two different types of drivers. In this case, an embodiment of machine learning engine 104 can advantageously be trained using detailed metadata about the electronic designs of the different types of drivers, making the updated machine learning engine sensitive to the consequences of electronic design choices on product reliability and / or product performance (such as carbon footprint over the product's lifetime). After the embodiment 104 has been trained with labeled failure data from different LED driver electronic designs, the updated machine learning engine can be used to predict the expected product lifetime, operating state, or maintenance requirements when considering variations in the LED driver's electronic design.

[0063] Thus, the exemplary development of arrangement 200 is suitable for closing the loop from product reliability insights to automated electronic design based on reliability insights generated by an updated machine learning engine, and for upgrading lighting device drivers with alternative, and often more expensive, components (e.g., different 3D-printed housings for better cooling, better capacitors). Electronic design upgrades can be location-specific; for example, for an LED driver to be used in Arizona, the model recommends a first type of capacitor, while for a streetlight project in Alaska, a second type of capacitor is recommended. Similarly, the luminaire housing can also be customized, for example, using different 3D-printed heat sinks.

[0064] In a further developed arrangement 200, each lighting fixture provider (eg, OEM) can add nighttime satellite imagery of the corresponding streetlight project to extract the health status of each installed lamp and the physical environment of each streetlight (including current lux levels).

[0065] In a further developed arrangement 200 , decision criteria are employed for what set of models to use (e.g. a first model for improving wireless communication of wireless drives or a second model for predicting instantaneous capacitor failures versus a third model tasked with predicting cumulative failures within the next five years).

[0066] In another exemplary arrangement, device status data (e.g., maintenance data) or a portion thereof is provided as a service to partners and, potentially, end customers. For example, a (QR) code certificate detailing the inference of a reliability model can be provided; for this (QR) code certificate to be useful in the used market, the remaining useful life prediction made by the updated machine learning engine needs to have proof of authenticity and provide confidence in the inference. Due to the disclosed SBLM-based collaboration across lighting device providers (e.g., manufacturers), the updated machine learning engine can learn from more labeled failures, resulting in a higher confidence in the remaining useful life inference than a single provider could ever achieve on its own. Since lighting devices only rarely fail, having insights into luminaire failures is extremely valuable.

[0067] Figure 2 A flowchart 300 is shown that includes exemplary training steps for swarm-based machine learning (SBLM) that can be implemented by the lighting device operational state estimation arrangement according to the present invention. Swarm-based learning is a decentralized machine learning (ML) solution that uses edge computing and is built on blockchain technology for peer-to-peer collaboration. Using a private permissioned blockchain, only the model structure and its training parameters (rather than the training dataset itself) are shared, which ensures the security and privacy of the data while still allowing everyone to benefit from collective learning. Data sovereignty, security, and privacy requirements can all pose obstacles to transmitting and aggregating the large amounts of data required to train complex ML models. In particular, swarm-based solutions have no single point of failure. All results of earlier learning are stored in the distributed ML engine, each of which has equal rights / capabilities.

[0068] In the first step 301, participating partners register a blockchain smart contract. In step 302, the ML engine's instantiation is received by all participating partners and individually trained using their respective proprietary training datasets. When the trigger conditions checked in step 303 are met, the participating partners output the current model parameters in step 304, and in step 305, these are sent to the swarm application programming interface (API). In step 306, a merged parameter set is obtained, and in step 307, the model is updated using the merged parameters. In step 308, a stopping criterion is checked. Depending on whether the stopping criterion is met, the entire process is stopped in step 309, or the updated model is further trained in step 302.

[0069] Figure 3A flow chart of a method 400 for estimating the operational status (such as maintenance requirements) of a target lighting fixture according to an embodiment of the present invention is shown. The method includes, in step 401, providing respective instantiations of a machine learning engine with an initial parameter set to a plurality of lighting fixture providers, each of which provides respective lighting fixtures for one or more lighting installations. The method also includes, in step 402, providing respective proprietary training datasets to the respective instantiations for training the machine learning engines, wherein the training datasets include state data and / or operational data of test lighting fixtures provided by the lighting fixture providers. The method also includes, in step 403, determining an updated parameter set indicative of the trained machine learning engine using the proprietary training dataset and the respective instantiations of the machine learning engine. The method also includes, in step 404, providing engine data indicative of the respective trained machine learning engine for generating an updated machine learning engine in step 405. The method also includes, in step 406, generating and providing device state data indicative of the operational status (such as maintenance requirements) of the target lighting fixtures using the updated machine learning engine and current device data including state data and / or operational data of the target lighting fixtures. Preferably, the device status data includes operational life data indicating an expected operational life of the target lighting device, and preferably, the engine data provided by the lighting device provider consists of an updated parameter set.

[0070] Method 400 may optionally include, in step 407, providing the updated machine learning engine to one or more of the plurality of lighting fixture providers and / or to one or more third-party lighting fixture providers. Optionally, the provided updated machine learning engine may be used as a materialization of the machine learning engine (such as the materialization provided in step 401) for further training and updating the machine learning engine. This iterative training may be performed until a predetermined stopping criterion is met. The stopping criterion is typically a reliability-based criterion, wherein a reliability threshold indicating the significance of the estimate must be exceeded before the iterative training of the ML engine ceases.

[0071] Preferably, method 400 further comprises, in step 408, encrypting engine data indicating the corresponding trained machine learning engine before providing the engine data for generating an updated machine learning engine, in particular providing encrypted updated engine parameters as engine data.

[0072] Additionally or alternatively, exemplary method 400 includes, in step 409 , storing device state data on a memory unit of the lighting device, in particular using blockchain technology.

[0073] Generally speaking, a pre-trained model, called a machine learning engine, based on a fixed model architecture is shared with lighting device providers (such as OEMs). Over time, the OEM feeds the model with proprietary training data, including luminaire metadata (luminaire type, GPS location, etc.), environmental data (including time-series weather data on storms and lightning strikes), and fault events and service actions. After locally retraining its ML model with its proprietary dataset, the OEM shares the updated ML engine parameters with other OEMs participating in the collective learning. Thus, the distributed SBML parameter set consolidates and aggregates all OEM learning from different luminaire suppliers regarding, for example, reliability or product performance over time, without requiring any sharing of actual project or customer data.

[0074] In summary, the present invention relates to a lighting device operating state estimation device for estimating the operating state of a target lighting device, which is configured to: provide corresponding instantiations of a machine learning engine with an initial parameter set to multiple lighting device providers; receive engine data indicating corresponding trained machine learning engines trained using corresponding proprietary training data sets provided by the multiple lighting device providers from at least a subset of the multiple lighting device providers; use the corresponding engine data to generate and provide an updated machine learning engine; and use the updated machine learning engine and current device data to generate and provide device state data indicating the operating state of the target lighting device with increased reliability.

[0075] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

[0076] In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" does not exclude a plurality.

[0077] A single unit or device may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0078] The computer program may be stored / distributed on a suitable medium, such as optical storage media or solid-state media, provided together with or as part of other hardware; but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0079] Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. A lighting device operating state estimation device (100) for estimating an operating state of a target lighting device (502), comprising: a machine learning engine providing unit (102) configured to provide respective instantiations of the machine learning engine (104) with an initial parameter set (P) to a plurality of lighting device providers (106, 108, 110), wherein each lighting device provider provides respective lighting devices (500a, 500b, 500c) for one or more lighting fixtures; an engine data receiving unit (112) configured to receive corresponding engine data (Ea, Eb, Ec) from at least a subset of the plurality of lighting device providers, the engine data including only or at least a portion of the updated parameter sets (Pa, Pb, Pc) derived from corresponding trained machine learning engines (104a, 104b, 104c), the corresponding trained machine learning engines (104a, 104b, 104c) being trained using corresponding proprietary training data sets (Ta, Tb, Tc), the corresponding proprietary training data sets (Ta, Tb, Tc) including status data and / or operation data of test lighting devices (500a, 500b, 500c) provided by the lighting device providers; a machine learning engine updating unit (114) configured to generate and provide an updated machine learning engine (105) from the machine learning engine (104) using the corresponding updated parameter set; and - an operating state estimation unit (116) configured to generate and provide device state data (M) indicating an operating state of the target lighting device using the updated machine learning engine (105) and current device data (D) including state data and / or operating data of the target lighting device (502). 2 . The lighting device operating state estimation device according to claim 1 , wherein the operating state estimation unit is further configured to generate and provide an operating instruction for operating the target lighting device using the device state data.

3. A lighting device operating state estimation arrangement (200), comprising: - The lighting device operating state estimation device (100) according to claim 1 or 2; and - a plurality of machine learning engine units (202, 204, 206) configured to: receive respective instantiations of the machine learning engine (104) with an initial parameter set from the lighting device operating state estimation device (100); Providing a corresponding proprietary training data set (Ta, Tb, Tc) for training a machine learning engine, wherein the corresponding proprietary training data set (Ta, Tb, Tc) includes status data and / or operation data of a test lighting device provided by a corresponding lighting device provider; Using a proprietary training data set and a corresponding embodiment of a machine learning engine, determining a corresponding updated parameter set (Pa, Pb, Pc) derived from the trained machine learning engine (104a, 104b, 104c); and providing the updated parameter set to a lighting device operating state estimation device (100).

4. A method (400) for estimating an operating state of a target lighting device (502), comprising the following steps: - providing (401) respective instantiations of the machine learning engine with the initial parameter set to a plurality of lighting fixture providers, each lighting fixture provider providing respective lighting fixtures for one or more lighting facilities; - each lighting fixture provider provides (402) a corresponding proprietary training data set to a corresponding embodiment for training the machine learning engine, wherein the training data set includes status data and / or operation data of the test lighting fixtures provided by the lighting fixture provider; - each lighting fixture provider uses the proprietary training dataset and the corresponding instantiation of the machine learning engine to determine (403) an updated set of parameters derived from the trained machine learning engine; - each lighting fixture provider provides (404) engine data indicative of a corresponding trained machine learning engine for generating (405) an updated machine learning engine; and - Using the updated machine learning engine and current device data including state data and / or operational data of the target lighting device, generating and providing (406) device state data indicative of an operational state of the target lighting device.

5. The method (400) of claim 4, wherein the device status data includes operational lifetime data indicative of an expected remaining useful lifetime of the target lighting device.

6. The method (400) of any one of the preceding claims 4 to 5, further comprising providing (407) the updated machine learning engine to one or more of the plurality of lighting device providers and / or to one or more third-party lighting device providers.

7. A method according to any one of claims 4 to 6, wherein the proprietary training data set includes status data and / or operational data indicating one or more of metadata of the lighting device, environmental data associated with environmental conditions at the operating location of the lighting device, fault data indicating operational faults of the lighting device, and service data indicating service actions performed on the lighting device.

8. The method of claim 7, wherein the lighting device metadata indicates one or more of a type of lighting device, a serial number of the lighting device, a production date of the lighting device, a production location of the lighting device, an operating location of the lighting device, a usage condition of the lighting device, lighting device components of the lighting device, and a driver type of the lighting device.

9. The method (400) according to claim 7 or 8, wherein the environmental data is indicative of weather conditions during operation of the lighting device at the installation location, and / or is indicative of conditions to which the lighting device has been exposed before its installation.

10. The method (400) of any one of the preceding claims 4 to 9, wherein the proprietary training data set further comprises image data or light sensor data indicative of lux levels of one or more lighting devices.

11. The method (400) according to any one of the preceding claims 4 to 10, further comprising the following steps: In particular, the device status data (409) is stored on a memory unit of the lighting device using blockchain technology.

12. Use of a machine learning engine or a lighting device operating state estimation arrangement according to claim 3, for performing the method for estimating the operating state of a target lighting device according to any one of the preceding claims 4 to 11.

13. A computer program comprising instructions which, when executed by a computing system of the lighting device operating state estimation arrangement according to claim 3, cause the computing system to perform the method according to any one of the preceding claims 4 to 11.

Images