System for industrial applications

Abstract

Industrial system (30) comprising input / output peripherals (31) configured for the implementation of a specific industrial application; an interface carrier board (20) comprising connectors (21) for the connection to said input / output peripherals (31); and an electronic board (10) comprising a CPU (14) and a dedicated artificial intelligence accelerator (12), distinct from the CPU (14), wherein the CPU (14) and the artificial intelligence accelerator (12) are connected to each other by means of data exchange interfaces. Specifically, on the artificial intelligence accelerator (12) and the CPU (14) there is installed at least one software package suitably configured for the implementation of said specific industrial application; the electronic board (10) is mechanically removably mounted on the carrier board (20) by means of connection members (11); the electronic board (10) and the carrier board (20) are connected to each other by means of data exchange interfaces (13, 23); and the interface carrier board (20) is connected to said input / output peripherals (31) by means of said data exchange connectors (21).

Description

[0001] "System for industrial applications"

[0002] DESCRIPTION

[0003] The present invention relates to a system for industrial applications.

[0004] The invention is also directed to a method for realizing an industrial system, to a method for controlling the operation of an industrial system as well as to an electronic device and a kit for making the industrial system.

[0005] In particular, the present invention finds application in an industrial context relating, for example, to the automotive sector, the road sector, the sector of smart homes and / or cities, the sector of the internet of things and the sector of industrial production plants.

[0006] The use of artificial intelligence in the industrial field is becoming increasingly widespread for various applications such as, for example, for the recognition / detection of people and / or objects, recognition of the position of people and / or objects, gesture recognition, command of industrial robots, anomaly detection, inspection operations, surveillance operations, monitoring operations, assisted and automated driving of vehicles and the like.

[0007] Compared to consumer goods intended for end consumers, electronic devices intended for an industrial context must in general provide high guarantees in terms of, for example, flexibility, customisation and reliability.

[0008] In this context, the Applicant has perceived the need to provide a solution that allows the use of artificial intelligence for specific industrial applications in a flexible, reconfigurable, customised and reliable way.

[0009] The Applicant has found that the aforementioned need can be met by a solution in which: the "smart" part for the execution of artificial intelligence algorithms is entrusted to dedicated electronic components (suitably, a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, arranged on an electronic board); the interface part to external input / output peripherals is entrusted to a carrier board whose main function is to interface with the electronic components and the input / output peripherals; and the "smart" part is configured to be removably mounted on the interface part. The present invention therefore relates to, in a first aspect thereof, to an industrial system.

[0010] Preferably, the industrial system comprises input / output peripherals configured for the implementation of a specific industrial application.

[0011] Preferably, the industrial system comprises an interface carrier board comprising data exchange connectors for the connection to said input / output peripherals.

[0012] Preferably, the industrial system comprises an electronic board comprising a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU.

[0013] Preferably, the CPU and the artificial intelligence accelerator are connected to each other by means of data exchange interfaces.

[0014] Preferably, at least one software package suitably configured for the implementation of said specific industrial application is installed on the artificial intelligence accelerator and on the CPU.

[0015] Preferably, the electronic board is removably mounted on the carrier board by means of mechanical connection members.

[0016] Preferably, the electronic board and the carrier board are connected to each other by means of data exchange interfaces.

[0017] Preferably, the interface carrier board is connected to said input / output peripherals by means of said data exchange connectors.

[0018] In a second aspect thereof, the present invention relates to a reconfigurable industrial system.

[0019] Preferably, the reconfigurable industrial system comprises input / output peripherals configured for the implementation of a specific industrial application.

[0020] Preferably, the reconfigurable industrial system comprises an interface carrier board comprising data exchange connectors for the connection to said input / output peripherals.

[0021] Preferably, the reconfigurable industrial system comprises electronic components used for the execution of artificial intelligence algorithms suitably configured for the implementation of said specific industrial application.

[0022] Preferably, said electronic components are arranged on an electronic board which is removably mounted on the carrier board by means of mechanical connection members. Preferably, the electronic board and the interface carrier board are connected to each other by means of data exchange interfaces;

[0023] Preferably, the interface carrier board is connected to said input / output peripherals by means of said data exchange connectors.

[0024] In a third aspect thereof, the present invention relates to a method for realizing an industrial system.

[0025] Preferably, in the method for realizing, it is provided for the provision of input / output peripherals configured for the implementation of a specific industrial application.

[0026] Preferably, in the method for realizing, it is provided for the provision of an interface carrier board comprising data exchange connectors for the connection to said input / output peripherals.

[0027] Preferably, in the method for realizing, it is provided for the provision of an electronic board comprising a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, wherein the CPU and the artificial intelligence accelerator are connected to each other by means of data exchange interfaces.

[0028] Preferably, in the method for realizing, it is provided for the provision, in a server, of a collection of software comprising a plurality of software packages, each software package comprising a series of software programs that are jointly configured for the installation on the artificial intelligence accelerator and on the CPU and for the implementation of a respective industrial application among a plurality of industrial applications.

[0029] Preferably, in the method for realizing, it is envisaged obtaining from the server at least one software package, among said plurality of software packages, suitably configured for the implementation of said specific industrial application among said plurality of industrial applications and installing said at least one software package thus obtained on the artificial intelligence accelerator and on the CPU.

[0030] Preferably, in the method for realizing, it is provided for removably mounting the electronic board on the carrier board, the electronic board and the carrier board being connected to each other by means of data exchange interfaces.

[0031] Preferably, in the method for realizing, it is provided for connecting the interface carrier board to said input / output peripherals by means of said data exchange connectors.

[0032] In a fourth aspect thereof, the present invention relates to a method for controlling the operation of an industrial system.

[0033] Preferably, in the control method, it is provided for the provision of input / output peripherals configured for the implementation of a specific industrial application.

[0034] Preferably, in the control method, it is provided for the provision of an interface carrier board comprising data exchange connectors for the connection to said input / output peripherals.

[0035] Preferably, in the control method, it is provided for the provision of an electronic board comprising a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, wherein the CPU and the artificial intelligence accelerator are connected to each other by means of data exchange interfaces.

[0036] Preferably, in the control method, it is provided for obtaining from a server at least one software package suitably configured for the implementation of said specific industrial application and to install said at least one software package thus obtained on the artificial intelligence accelerator and on the CPU.

[0037] Preferably, in the control method, it is provided for removably mounting the electronic board on the carrier board, the electronic board and the carrier board being connected to each other by means of data exchange interfaces.

[0038] Preferably, in the control method, it is provided for connecting the interface carrier board to said input / output peripherals by means of said data exchange connectors.

[0039] Preferably, in the control method, it is provided for operating the input / output peripherals by running said at least one software package installed on the artificial intelligence accelerator and on the CPU.

[0040] In a fifth aspect thereof, the present invention relates to an electronic device.

[0041] Preferably, the electronic device comprises an interface carrier board to input / output peripherals configured for the implementation of an industrial application, the carrier board comprising data exchange connectors for the connection to said input / output peripherals. Preferably, the electronic device comprises an electronic board comprising a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, the CPU and the artificial intelligence accelerator being connected to each other by means of data exchange interfaces;

[0042] Preferably, the electronic board and the interface carrier board are provided with data exchange interfaces and the electronic board comprises mechanical connection members configured to removably mount the electronic board on the carrier board.

[0043] Preferably, the electronic board is mounted on the carrier board by means of said connection members and interfaces with the carrier board by means of said data exchange interfaces.

[0044] In a sixth aspect thereof, the present invention relates to a kit for industrial systems.

[0045] Preferably, the kit comprises an electronic board comprising a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, the CPU and the artificial intelligence accelerator being connected to each other by means of data exchange interfaces.

[0046] Preferably, the kit comprises a collection of software stored in a server and comprising a plurality of software packages, each software package comprising a series of software programs that are jointly configured for the installation on the artificial intelligence accelerator and on the CPU and for the implementation of a respective industrial application among a plurality of industrial applications.

[0047] Preferably, the electronic board comprises mechanical connection members configured to removably mount the electronic board on an interface carrier board to input / output peripherals configured for the implementation of a specific industrial application among said plurality of industrial applications.

[0048] Preferably, the electronic board comprises a data exchange interface between the electronic board and the carrier board.

[0049] In a seventh aspect thereof, the present invention relates to an industrial system for controlling a drone.

[0050] Preferably, the industrial system comprises input / output peripherals. Preferably, the industrial system comprises an interface carrier board comprising data exchange connectors for the connection to said input / output peripherals.

[0051] Preferably, the industrial system comprises an electronic board comprising a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU.

[0052] Preferably, the CPU and the artificial intelligence accelerator are connected to each other by means of data exchange interfaces.

[0053] Preferably, at least one suitably configured software package controlling a drone is installed on the artificial intelligence accelerator and on the CPU.

[0054] Preferably, the electronic board is removably mounted on the carrier board by means of mechanical connection members.

[0055] Preferably, the electronic board and the carrier board are connected to each other by means of data exchange interfaces.

[0056] Preferably, the interface carrier board is connected to said input / output peripherals by means of said data exchange connectors.

[0057] Preferably, the artificial intelligence accelerator and the CPU are configured to control the drone based on an input received from at least one of said input / output peripherals.

[0058] Preferably, the control of the drone by means of the industrial system takes place in real time.

[0059] Thanks to the various aspects of the invention, having an electronic board comprising both a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, it is possible to reliably and effectively implement industrial applications that make use of artificial intelligence and require high and dedicated computing capabilities. Furthermore, by providing a "general purpose" electronic board in which the CPU and the artificial intelligence accelerator are preconfigured to interact with each other, the invention provides a ready-to-use solution, in which the electronic board is ready to host, from time to time, depending on the specific industrial application of interest, a software package suitably configured for the implementation of the specific industrial application among a plurality of possible industrial applications. In addition, having an interface carrier board available, the electronic board thus configured can be mounted on this carrier board and then connect at least part of the connectors of the carrier board to input / output peripherals useful for the implementation of the specific industrial application.

[0060] An operator interested in creating an industrial system for a specific industrial application (for example relating to the control of industrial robots) has the possibility of implementing this application by preparing the appropriate input / output peripherals (for example, sensors, displays, industrial robots...), a special interface carrier board equipped with connectors suitable for the connection with such input / output peripherals and a "general purpose" electronic board, equipped with CPU and artificial intelligence accelerator, suitable for being mounted on the carrier board. After that, the operator has the possibility to adapt the "general purpose" electronic board to the implementation of the specific industrial application by obtaining, from a specific server, a software package suitably pre-configured for the implementation of said specific industrial application and installing said software package on the artificial intelligence accelerator and on the CPU. Considering that the artificial intelligence accelerator and the CPU are already pre-configured to interact with each other and the software package comprises a series of software programs that are jointly configured for the installation on the artificial intelligence accelerator and on the CPU and for the implementation of the specific industrial application, the configuration operations of the industrial system are extremely simple and immediate.

[0061] It is also noted that, in the event that the input / output peripherals are equipped with standard connectors, it will be possible to mount the electronic board on a "general purpose" carrier board, equipped with standard connectors and, possibly, use - among these standard connectors - only those necessary for the specific implementation. On the other hand, in the event that the input / output peripherals are equipped with special connectors, it will be possible to mount the electronic board on a specific interface carrier board, equipped with special connectors suitable for the specific use.

[0062] In addition, the fact that the electronic board can be removably mounted on the interface carrier board makes the solution particularly flexible and easily adaptable to different contexts, needs and applications. For example, if it is necessary to add new peripherals or replace existing peripherals with others with connectors different from those present in the interface carrier board, it will be possible to disassemble the electronic board from the current carrier board to mount it on a new carrier board equipped with the new connectors required. In addition, if it is necessary to replace the artificial intelligence accelerator and / or the CPU with others with different characteristics (for example having greater computing capability) than those present in the current electronic board, it will be possible to remove the current electronic board from the carrier board to mount on the same carrier board a new electronic board equipped with the new artificial intelligence accelerator and / or the new CPU.

[0063] As a whole, the invention in its various aspects therefore provides a flexible, reconfigurable, reliable and customizable solution for the realization of industrial applications that make use of artificial intelligence.

[0064] Throughout this description, the terms "artificial intelligence accelerator" and "Al accelerator" are used interchangeably and have the same meaning.

[0065] An artificial intelligence accelerator, also called a deep learning processor or a neural processing unit (NPU), is intended to refer to a class of specialized hardware accelerators or computer systems designed to accelerate applications of artificial intelligence and machine learning, including artificial neural networks and machine vision. Generally, an Al accelerator is capable of processing at least 0.5 TOPS (tera operations per second), preferably tens or hundreds of TOPS.

[0066] The term "dedicated artificial intelligence accelerator, distinct from the CPU" is intended to mean an artificial intelligence accelerator not integrated into the CPU chip, i.e. an artificial intelligence accelerator made in a chip other than the chip in which the CPU is made. Specifically, the chip in which the artificial intelligence accelerator is made and the chip in which the CPU is made are mounted on the same electronic board.

[0067] By the term "real-time", it is meant a low latency process.

[0068] Preferred features of the various aspects of the invention are described below, each of these features being capable of being provided individually or in combination with the others.

[0069] In a preferred embodiment, the electronic board has a compact form factor with the three sizes respectively less than 60 mm x 50 mm x 5 mm, where the sizes indicated are to be understood with a tolerance of about 5%.

[0070] Preferably, the carrier board is devoid of electronic components having computing capability suitable for the execution of artificial intelligence algorithms. Specifically, the carrier board is devoid of hardware, firmware and software suitable for the execution of artificial intelligence algorithms.

[0071] Preferably, the carrier board is devoid of Al accelerator.

[0072] Preferably, the main function of the carrier board is to interface with the input / output peripherals and with the electronic board.

[0073] However, in other embodiments, the carrier board may comprise electronic components having computing capabilities suitable for the execution of artificial intelligence algorithms dedicated to the execution of specific operations such as, for example, the monitoring and supervision of input / output peripherals.

[0074] In a preferred embodiment, the electronic board is devoid of an internal fan. This is possible when using low-power CPU and Al accelerator.

[0075] Preferably, the CPU and the artificial intelligence accelerator are connected to each other by means of a PCI Express type data exchange interface.

[0076] The electronic board may also comprise a GPU (graphics processing unit) distinct from the Al accelerator.

[0077] The GPU can be distinct from the CPU or integrated into the CPU.

[0078] Preferably, the electronic board comprises at least one volatile memory (for example RAM).

[0079] Preferably, the electronic board comprises at least one non-volatile memory.

[0080] In other words, preferably, the electronic board comprises at least one memory. The at least one memory may preferably be a volatile and / or non-volatile memory.

[0081] Preferably, the at least one memory and the artificial intelligence accelerator are connected to each other by means of data exchange interfaces.

[0082] Preferably, the at least one memory is connected so as to exchange data with the artificial intelligence accelerator.

[0083] Preferably, the at least one memory and the CPU are connected to each other by means of data exchange interfaces.

[0084] Preferably, the at least one memory is connected so as to exchange data with the CPU.

[0085] Preferably, the at least one memory is connected to the artificial intelligence accelerator by means of a PCI express interface.

[0086] Preferably, at least one memory is configured to receive a Large Language Model. The at least one memory may preferably be a volatile and / or non-volatile memory.

[0087] Preferably, the Large Language Model is configured for language recognition.

[0088] Preferably, the Large Language Model is executed by the artificial intelligence accelerator.

[0089] Preferably, the artificial intelligence accelerator is configured to run the Large Language Model to recognize a language and / or a text.

[0090] Preferably, the artificial intelligence accelerator is configured to run the Large Language Model to provide information by processing the language and / or text.

[0091] Preferably, the input / output device is connected to a microphone-type transducer. Preferably, the artificial intelligence accelerator is for executing the Large Language Model that can process the language coming from the microphone.

[0092] Preferably, the electronic board is configured to convert the language into data processable by the artificial intelligence accelerator and / or the CPU.

[0093] The electronic board can be powered by the interface carrier board (for example for low consumption applications) and / or by a dedicated power supply external to the electronic board and the carrier board (for example for high consumption applications).

[0094] The electronic board is preferably connected to the input / output peripherals indirectly, by means of the data exchange connectors between the interface carrier board and the input / output peripherals.

[0095] In addition or alternatively, the electronic board can be connected at least in part to the input / output peripherals directly, through suitable connectors adapted to allow a data exchange directly between the electronic board and the input / output peripherals.

[0096] In one embodiment, the CPU comprises a further Al accelerator integrated therein, distinct from the aforesaid Al accelerator present on the electronic board.

[0097] In the reconfigurable industrial system according to the second aspect of the invention, the electronic components comprise a CPU and a dedicated artificial intelligence accelerator, distinct from the CPU, the CPU and the artificial intelligence accelerator being connected to each other by means of data exchange interfaces.

[0098] The data exchange connectors between the interface carrier board and the input / output peripherals may be of the wired or wireless type (for example wireless or Bluetooth).

[0099] When present, the data exchange connectors between the electronic board and the input / output peripherals can be of the wired or wireless type (for example wireless or Bluetooth).

[0100] Preferably, the data exchange connectors between the interface carrier board and the input / output peripherals and any data exchange connectors between the electronic board and the input / output peripherals comprise at least one connector selected from the group comprising: HDMI connectors, USB connectors, Ethernet connectors, serial connectors (for example of the RS-232 and RS-485 type), GPI connectors, GPO connectors, I2C connectors, SPI connectors, PCI connectors, PCIe connectors, memory card connectors, external fan connector, MIPI-DSI connectors, MIPI-CSI connectors, SDIO connectors, PCM connectors, PWM connectors, Canbus connectors.

[0101] Preferably, the aforesaid data exchange connectors are configured for the connection to at least one input / output peripheral selected from the group comprising: a sensor or other data source as input, a control device, a transducer, a display, a computer, an industrial robot, a memory medium, a fieldbus.

[0102] In other words, preferably, the electronic board is configured to communicate with multiple input / output peripherals which are in turn placed on the carrier board.

[0103] Preferably, the electronic board is configured to receive data as input by means of one or more input / output peripherals. Preferably, the electronic board is configured to provide data as output by means of one or more input / output peripherals. Preferably, the input / output device is selected from the group comprising: a sensor, a control device, a display, a transducer, a computer, an industrial robot, a memory medium, a fieldbus. In other words, preferably, at least two input / output peripherals may be of at least two different types from each other.

[0104] Preferably, the industrial application is related to at least one sector selected from the group comprising the automotive sector, the road sector, the smart homes and / or cities sector, the internet of things sector, the industrial production plant sector, the automation sector, the robotics sector, the drone sector, the quality control sector, the 3D vision systems sector, the stereovision sector, the smart camera sector and the like.

[0105] Preferably, the industrial application is related to at least one application selected from the group comprising: recognition and / or detection of people and / or objects, recognition of the position of people and / or objects, 3D pose estimation, 3D depth estimation, gesture recognition, command of industrial robots, detection of anomalies of operation of a device or plant, control of operation of a device or plant, detection of anomalies in a product, surveillance operations, monitoring operations (for example of traffic), assisted and automated driving operations of vehicles and the like.

[0106] In one embodiment, the input / output peripherals comprise at least one camera and an industrial robot. In this embodiment, said at least one software package installed on the artificial intelligence accelerator and on the CPU is preferably configured to process images captured by the camera and control the industrial robot based on the result of processing said images. The image processing can, for example, be carried out (for example by implementing Deep Learning algorithms) by the artificial intelligence accelerator while the command of the industrial robot can be carried out by the CPU. Preferably, the camera is configured to capture the movements (for example of the hand) of an operator and said at least one software package installed on the artificial intelligence accelerator and on the CPU is preferably configured to recognize the movements of the operator by processing the images captured by the camera and command the industrial robot based on the result of processing said images. Preferably, in this embodiment, the input / output peripherals also comprise a screen for providing the operator with feedback on his / her movements and / or on the relationship between his / her movements and those of the industrial robot.

[0107] In a further embodiment, the input / output peripherals comprise at least one camera and a driver assistance system for applications in the automotive sector. In this embodiment, said at least one software package installed on the artificial intelligence accelerator and on the CPU is preferably configured to process images captured by the camera and operate the driver assistance system on the basis of the result of the processing of said images. Image processing may, for example, be performed (for example, by implementing Deep Learning algorithms) by the artificial intelligence accelerator while operation of the driver assistance system may be performed by the CPU.

[0108] In an embodiment for controlling the flight of a drone, the input / output peripherals preferably comprise a vision system.

[0109] Preferably, the vision system comprises one or more cameras.

[0110] Preferably, the vision system is configured to capture movements (for example of the hand) of an operator.

[0111] Preferably, the at least one software package installed on the artificial intelligence accelerator and on the CPU is configured to process images captured by the vision system and command the drone on the basis of the result of the processing of said images.

[0112] Preferably, the image processing is carried out by the artificial intelligence accelerator. Preferably, the control of the flight of the drone is carried out by the CPU.

[0113] Preferably, the industrial system is configured to detect a presence of at least one operator to authorize an interaction with the drone, and signals the activation with an LED or a visual signal shown to the operator.

[0114] Preferably, the vision system comprises a first camera mounted on board the drone. In this embodiment, the industrial system is preferably mounted on board the drone.

[0115] Preferably, the first camera comprises a night vision and / or infrared system. Preferably, the industrial system is configured to control and / or move the first camera for operator tracking.

[0116] In an alternative, preferably, the vision system comprises a first camera distinct from the drone. In this alternative, the industrial system is preferably positioned together with the vision system, for example on the ground, and the input / output peripherals comprise a radio module for sending commands to the drone. In both of the above cases, preferably, the first camera is configured to provide images of operator gestures. Preferably, the industrial system is configured to command the drone based on said gestures.

[0117] Advantageously, having the first camera and the industrial system mounted on board the drone makes it possible to create an autonomous vehicle that integrates all the elements necessary for its control, with greater portability and interaction efficiency.

[0118] Advantageously, in the alternative of the first camera and the industrial system mounted on the ground for a remote control, it is possible to create a lighter drone; thanks to the radio module for the control, moreover, the drone does not need to remain within sight with the operator.

[0119] Preferably, the industrial system is configured for a recognition of objects framed by the vision system, and for an automatic landing of the drone.

[0120] Preferably, the input / output peripherals comprise a second camera mounted onboard the drone. Preferably, the second camera is configured for operational functions of the drone, such as inspections.

[0121] In an embodiment for controlling the flight of a drone, the input / output peripherals preferably comprise an audio peripheral, such as a microphone, as an alternative or in addition to the vision system.

[0122] Preferably, the audio device is configured to acquire an operator language. Preferably, the audio device is associated with the industrial system.

[0123] Preferably, the audio device is mounted onboard the drone. Alternatively, preferably, the audio device is distinct from the drone, for example on the ground together with the vision system.

[0124] Preferably, the at least one software package installed on the artificial intelligence accelerator and on the CPU is configured to process, by means of a Large Language Model, the language of the operator and command the drone on the basis of the result of the language processing.

[0125] Preferably, the industrial system is configured to command the drone based on voice commands given by the operator.

[0126] Preferably, the industrial system is configured to select a predominant command mode between the voice commands and the operator gestures.

[0127] Preferably, the selection of the predominant control mode is carried out based on a processing of the artificial intelligence accelerator.

[0128] Preferably, in this embodiment, the input / output peripherals may comprise a screen for providing the operator with feedback on his / her movements and / or language and / or on the relationship between his / her commands and the flight of the drone.

[0129] In a further embodiment of any of the various aspects of the invention, it is possible to provide a further electronic board having the features described above.

[0130] This additional electronic board can be removably mounted on the same carrier board as the aforementioned electronic board and be configured for data exchange with the aforementioned electronic board. The aforementioned electronic board and the additional electronic board can be configured to interact with different input / output peripherals in a coordinated manner.

[0131] Such a solution with a further electronic board may, for example, be advantageous in case it is necessary to enhance the computing capability to interact with the various input / output peripherals and / or to provide a redundant system in which particular safety-critical functions are performed by both electronic boards, in particular by the two Al accelerators and / or by the two CPUs of the two electronic boards.

[0132] Alternatively, it may be provided for mounting the further electronic board on a further carrier board. In this case, each carrier board will be connected to respective input / output peripherals. In addition, the two carrier boards and the two electronic boards will be configured to interact with each other so as to interact with the various input / output peripherals in a coordinated manner with each other.

[0133] In other words, preferably, the industrial system comprises at least one carrier board. Preferably, the industrial system comprises two or more carrier boards. Preferably, the two or more carrier boards each comprise at least one respective electronic board. Preferably, the at least one respective electronic board is removably mounted. Preferably, the two or more carrier boards are connected to each other by means of second data exchange interfaces. Preferably, the second data exchange interfaces are input / output peripherals. Preferably, the second data exchange interfaces are connected to one or more data exchange connectors of the respective carrier board. Preferably, the two or more carrier boards are configured to communicate with each other.

[0134] In other words, preferably, the industrial system comprises at least one electronic board. Preferably, the at least one respective electronic board is removably mounted on the carrier board. Preferably, the industrial system comprises two or more electronic boards. Preferably, the two or more electronic boards are removably mounted on the carrier board. Preferably, the two or more electronic boards are connected to each other by means of the data exchange interfaces. Preferably, the data exchange interfaces through which the two or more electronic boards are connected to each other are input / output peripherals. Preferably, the two or more electronic boards on the same carrier board are configured to communicate with each other.

[0135] Further features and advantages of the present invention will become clearer from the following detailed description of preferred embodiments, with reference to the appended drawings and provided by way of indicative and non-limiting examples, in which:

[0136] Figure 1 schematically shows an electronic device according to an embodiment of the invention;

[0137] - Figure 2 schematically shows an industrial system according to an embodiment of the invention;

[0138] - Figure 3 schematically shows an example of the industrial system of Figure 2;

[0139] - Figure 4 shows a first example of industrial application in which an industrial system according to the invention can be used;

[0140] - Figure 5 shows a second example of industrial application in which an industrial system according to the invention can be used.

[0141] Figure 1 shows an electronic device 1 according to an embodiment of the invention that allows artificial intelligence to be used in industrial applications.

[0142] The device 1 comprises an interface carrier board 20 and an electronic board 10.

[0143] The interface carrier board 20 comprises connectors (collectively indicated in the figure with reference numeral 21 ) for the connection to input / output peripherals 31.

[0144] The connectors 21 are configured to allow an exchange of data and possibly an electric power transmission.

[0145] The connectors 21 may be of the wired or wireless type (for example wireless or Bluetooth).

[0146] The connectors 21 may be selected from the group comprising: HDMI connectors, USB connectors, Ethernet connectors, serial connectors (for example RS-232 and RS-485 type), GPI connectors, GPO connectors, I2C connectors, SPI connectors, PCI connectors, PCIe connectors, memory card connectors, external fan connector, MIPI-DSI connectors, MIPI-CSI connectors, SDIO connectors, PCM connectors, PWM connectors, Canbus connectors.

[0147] Preferably, the electronic board 10 has a compact form factor with the three sizes (length, width and height) of substantially 55 mm x 40 mm x 4.7 mm, where the indicated sizes are to be understood with a tolerance of about 5%.

[0148] The electronic board 10 comprises a CPU 14 and a dedicated Al accelerator 12, distinct from the CPU 14.

[0149] Preferably, the electronic board 10 has expandable memory and MIPI-DSI and MIPI-CSI connectors.

[0150] Preferably, the CPU 14 is of the multicore type with a dedicated operating system (for example Linux). The Al accelerator 12 is a processor capable of processing at least tens or hundreds of TOPS and has a power in the 1 -50W range. Preferably, the CPU 14 is realized in a microprocessor.

[0151] For example, the CPU 14 may be a processor of the i.MX8M™ Plus family from the company NXP or a processor from the company Rockchip, such as a RK3588 model processor.

[0152] The Al accelerator 12 may be a Hailo-8™ processor from the company Hailo or a processor from the company Axelera.

[0153] In the illustrated embodiment, the electronic board 10 also comprises a volatile memory 15 (for example a RAM) and a non-volatile memory 16. The CPU 14 and the Al accelerator 12 are provided with appropriate interfaces (schematically illustrated with a double arrow between the CPU 14 and the Al accelerator 12 in the figure) that allow data exchange and interoperability between them.

[0154] The electronic board 10 and the interface carrier board 20 are provided with data exchange interfaces 13, 23 that allow interoperability with each other when the electronic board 10 is mounted on the interface carrier board 20.

[0155] The interfaces 13, 23 may be configured to allow a transmission of electrical power, preferably from the interface carrier board 20 to the electronic board 10. In particular, depending on the specific application, the electronic board 10 may be powered by the interface carrier board 20 (for example for low-power applications) and / or by a dedicated external power supply (for example for high- power applications).

[0156] Furthermore, the electronic board 10 comprises mechanical connection members 11 configured to removably mount the electronic board 10 on the carrier board 20.

[0157] For example, the connection members 1 1 may comprise screws, self-locking pins or plastic and / or metal spacers.

[0158] Although not illustrated in Figure 1 , the electronic board 10 may also comprise a GPU distinct from the Al accelerator 12. Such a GPU may be distinct from the CPU 14 or integrated into the CPU 14.

[0159] The CPU 14 may comprise a further Al accelerator (not shown) integrated therein, distinct from the Al accelerator 12.

[0160] In a variant (not depicted) of the embodiment of figure 1 , the electronic board 10 comprises, in addition to the Al accelerator 12 and the CPU 14, a memory element dedicated to the implementation of a Large Language Model, positioned on the same electronic board 10.

[0161] An additional connector, of type CM4, is also provided for the electronic board 10.

[0162] Such dedicated memory element may be a portion of the volatile memory 15 (for example a RAM) or a portion of the non-volatile memory 16. Alternatively, the dedicated memory element may be a dedicated and distinct memory, both volatile (for example RAM) and non-volatile.

[0163] The dedicated memory element allows a sizeable Large Language Model to be loaded, directly executed on the Al accelerator 12, guaranteeing high performance.

[0164] The electronic board 10 comprises a PCI Express type connection between the CPU 14 and the Al accelerator 12, connecting with a dedicated bus the dedicated memory element for the Large Language Model.

[0165] In this implementation, the Applicant provides using, as an Al accelerator 12, a Hailo-H10™ processor from the company Hailo suitable for the implementation of deep neural networks, including Large Language Model, with very low power consumption. Alternatively, the Applicant provides using, as an Al accelerator 12, an Ara-2™ processor from the company Kinara, suitable for the implementation of Large Language Models with scalable architecture to manage particularly complex Al models.

[0166] In addition, the Applicant provides using for the CPU 14 a processor of the i.MX95™ family from the company NXP, based on arm Cortex-A55 and Cortex- M33 cores, particularly suitable for embedded applications with advanced security; this processor is equipped with versatile interfaces, supporting: PCI Express, Ethernet TSN, CAN-FD.

[0167] This variant, provided on a single electronic board 10: Al accelerator 12, CPU 14 and memory dedicated to the implementation of Large Language Model, is very effective for industrial applications that require high Al performance and operating stability.

[0168] Figure 2 schematically shows an industrial system 30 according to an embodiment of the invention.

[0169] The industrial system 30 comprises an electronic device 1 according to the invention for the functional and structural features of which reference is made to what has already been described above.

[0170] Furthermore, the industrial system 30 comprises a plurality of input / output peripherals 31 which are connected to the device 1 by means of the connectors 21 of the interface carrier board 20.

[0171] The input / output device 31 may be a sensor, a transducer, a control device, a lighting device, a display, a computer (for example, a PC, a server, a laptop, and the like), a memory medium, a fieldbus, and / or an industrial robot.

[0172] The sensor may be a video camera, a camera, a motion sensor, a weight sensor, a proximity sensor, a photoelectric sensor, a contour sensor, a displacement sensor, a position sensor, a CMOS sensor, a GPS sensor, a temperature sensor, a pressure sensor, and the like.

[0173] The transducer may be, for example, a microphone or a speaker.

[0174] The control device may comprise any regulation and / or control device such as, for example, an adjustable valve, a pressure regulator, a temperature regulator, a voltage regulator, a flow regulator, a switch, a relay as well as a complex decision-making system such as car control units, traffic managers, energy distribution systems etc.

[0175] By way of example, Figure 3 shows the industrial system of Figure 2 in which the input / output peripherals comprise a camera 31 a, an industrial robot 31 b, a lighting device 31 c and a display 31 d.

[0176] The industrial system 30 also comprises a remote server 40 accessible by means of a suitable communication network which in Figures 2 and 3 is schematically represented by a cloud 41 .

[0177] A collection of software comprising a plurality of software packages is stored in the server 40, wherein each software package comprises a series of software programs that are jointly configured for the installation on the Ai accelerator 12 and the CPU 14 and for the implementation of a respective industrial application among a plurality of industrial applications.

[0178] For example, the industrial application may relate to the automotive sector, the road sector, the smart home and / or smart city sector, the internet of things sector, the industrial manufacturing plant sector, the automation sector, the robotics sector, the drone sector, the quality control sector, the 3D vision systems sector, the stereovision sector, the smart camera sector and the like.

[0179] Specifically, the industrial application can be related to the recognition and / or detection of people and / or objects, the recognition of the position of people and / or objects, the estimation of 3D laying, the estimation of 3D depth, gesture recognition, voice recognition, the command of industrial robots, the detection of anomalies in the operation of a device or plant, the control of the operation of a device or system, the detection of anomalies in a product, surveillance operations, monitoring operations (for example of traffic), assisted and automated driving operations of vehicles and the like.

[0180] In order to create an industrial system 30 such as the one illustrated in Figure 2 or Figure 3 for a specific industrial application, it will be necessary to:

[0181] - provide the appropriate input / output peripherals 31 (for example, the camera 31 a, the industrial robot 31 b, the lighting device 31 c and the display 31 d) configured for the implementation of the specific industrial application;

[0182] - provide the interface carrier board 20 comprising connectors 21 for the connection to said input / output peripherals 31 ;

[0183] - provide the electronic board 10;

[0184] - obtain from the server 40 one or more software packages suitably configured for the implementation of said specific industrial application and install the software package or software packages thus obtained on the Al accelerator 12 and on the CPU 14 of the electronic board 10;

[0185] - removably mount the electronic board 10 on the carrier board 20 by means of the connection members 1 1 , the electronic board 10 and the carrier board 20 being connected to each other by means of the interfaces 13, 23 configured to allow interoperability between them;

[0186] - connect the interface carrier board 20 to the input / output peripherals 31 by means of the respective connectors 21 .

[0187] At this point, it will be possible to operate the input / output peripherals 31 by running the software package(s) installed on the Al accelerator 12 and on the CPU 14 of the electronic board 10.

[0188] By way of example, Figure 4 shows an industrial robot 31 b used by the Applicant to create an industrial system 30 such as that illustrated in Figure 3 to control the operation of the industrial robot 31 b in a "pick & place" application useful for picking products 32 from one position (for example, a tray 33) and repositioning them in another position (not illustrated in the figure). Although not illustrated, the industrial robot 31 b is equipped with an integrated vision system based on sensors to identify the products 32 to be picked and placed in the assigned position.

[0189] The industrial system 30 used to control the operation of the industrial robot 31 b is therefore able to support industrial operators in the selection and handling of objects in a "pick & place" environment.

[0190] By combining vision system, Al accelerator 12, and edge computing for CPU 14, industrial system 30 is able to recognize objects, provide operator assistance, and ensure a more efficient and error-free workflow.

[0191] In this example of operation of the industrial system 30, it is provided for a phase of acquisition and recognition of the object in which: the operator picks an object from a picking area; a vision system, such as an RGB and / or depth camera acquires the image of the object in real time; the industrial system 30 performs a detection and classification of the object, identifying the product and verifying that it is the correct one.

[0192] In the example of operation of the industrial system 30, it is provided for a validation and visual assistance phase, in which: if the selected object is correct, the industrial system 30 confirms visually and / or with an acoustic signal to the operator; if the object is incorrect, a visual and audible error is signalled and the operator is guided in the correct choice; on a display or on an augmented reality interface, the industrial system 30 can show additional information, such as: name and product code, position of the correct object in the warehouse, instructions for positioning or assembly.

[0193] In the example of operation of the industrial system 30, it is provided for a process traceability and optimization phase, in which: the industrial system 30 records each "picking" operation to monitor performance and identify any process bottlenecks; in addition, the industrial system 30 can be integrated with "warehouse management" systems (WMS) to update an inventory in real time.

[0194] In this implementation, the Applicant used, as Al accelerator 12, a Hailo-8™ processor from the company Hailo suitable for the implementation of Deep Learning algorithms for the detection of objects and estimation of their placement, while for CPU 14 used a processor of the i.MX8M™ Plus family from the company NXP for the management of the acquisition of images of products 32 (at a rate of at least 5 frames per second), for the control of the industrial robot 31 b and to delegate Deep Learning operations to the Hailo-8™ processor. As a further example, Figure 5 shows an application implemented by the Applicant with an industrial system 30 such as that illustrated in Figure 3 for controlling the operation of an industrial robot 31 b by recognizing gestures of an operator's hand.

[0195] In particular, in the implemented application, it is provided for recognising the presence of a face positioned in front of the camera 31 a for more than a certain number of seconds. Once the presence of the face is recognized, it is provided for turning on a light 31 c and to capture, through the camera 31 a, the gestures of the operator's hands in real time so as to control the movement of the industrial robot 31 b.

[0196] In particular, left hand gestures are used to control the movement of the arm 35 of the industrial robot 31 b while right hand gestures are used to control the movement of the gripping member 34 of the industrial robot 31 b. A screen 31 d shows the recognized face and hand gestures in real time.

[0197] In this implementation, the Applicant used, as an Al accelerator 12, a Hailo-8™ processor of the Hailo company suitable for the implementation of Deep Learning algorithms for the recognition of hand gestures and their position (at a speed of at least 30 frames per second), while for the CPU 14 used a processor of the i.MX8M™ Plus family of the NXP company for the recognition of the operator's face, for the control of the industrial robot 31 b and to delegate Deep Learning operations to the Hailo-8™ processor.

[0198] As a further example, the Applicant has implemented an industrial system 30 to control the flight of a drone in real time, by means of Al vision and processing systems on board the system. The flight of the drone may comprise take-off and / or cruise and / or landing and / or operational manoeuvres and / or emergency manoeuvres. In particular, in the implemented application, it is provided for recognizing objects, faces and gestures to control the drone without the use of a traditional remote control.

[0199] In this example of operation of the industrial system 30, it is provided for an operator recognition phase, in which: a vision system comprising at least one camera managed by the industrial system 30, detects the presence of the operator by recognizing for example his / her face, for example by comparing it with a database of predetermined faces; after a predefined recognition time, the industrial system 30 authorizes the interaction with the drone and signals the activation with an LED or a visual signal shown to the operator. For example, it can provided for an operator changeover phase in which: the industrial system 30 detects the presence of a second operator by recognising for example the face, for example by comparing it with a database of predetermined faces; after a predefined recognition time, the industrial system 30 disables the interaction of the first operator with the drone, and authorises the interaction of the second operator with the drone, possibly signalling the activation with an LED or a visual signal shown to the operator.

[0200] The vision system may comprise one or more cameras. In one example, the vision system comprises an operating camera of the drone itself, for example for inspections, and a camera designed to frame the operator to receive commands. At least one of the cameras may be a night vision camera or an infrared camera.

[0201] In the example of operation of the industrial system 30, it is provided for a visual tracking phase, in which: by means of a specific hand gesture, the operator can command the flight of the drone; a camera of the drone continues to track the operator, keeping the drone oriented towards him / her; if the face of the operator comes out of the frame of the camera, the drone can stop the movement or land automatically.

[0202] Alternatively, the camera designed to frame the operator to receive commands is a remote camera, positioned for example on the ground on a tripod and capable of sending commands to the drone. In this example, the industrial system 30 may be on the ground and connected to the camera. In this example, the industrial system 30 is configured to send operational commands to the drone, for example by means of a radio interface.

[0203] In the example of operation of the industrial system 30, it is provided for a realtime gesture control phase, in which: the movements of the operator's left hand are recognized by the camera and control the rotation and translation of the drone (right / left, forward / backward), while the movements of the operator's right hand are recognized by the camera and control the height and speed of the drone; the gestures are detected and interpreted in real time thanks to Deep Learning algorithms accelerated by the Al accelerator 12 with minimum latency.

[0204] In the example of operation of the industrial system 30, it is possible to provide, in addition to the vision system already described, an audio input device, for example a microphone associated with the operator. For example, the drone may comprise the microphone, for example using a directional microphone. Alternatively, for example, the industrial system 30 is distinct from the drone and grounded, and the microphone is a peripheral connected to the industrial system 30.

[0205] In this way, the operator can impart commands to the drone by gesture control, as described above, and in addition or alternatively can impart voice commands. For this purpose, the industrial system 30 comprises, in addition to the CPU 14 and the Al accelerator 12 arranged on the electronic board 10, a memory arranged on the electronic board 10 and configured to receive a Large Language Model or LLM. The Al accelerator 12 and / or the CPU 14 is configured to execute the LLM model to recognize an operator language. Such memory is for example a RAM memory.

[0206] The industrial system 30 is configured to receive information from the audio input device or microphone, and to recognise operator instructions to be applied to drone control. In this way, the operator can command with simple voice commands such as "right" / "left", "forward " / "backward" the rotation and translation of the drone, as well as command with simple voice commands such as "raise" / "lower" or "fast" / "slow down" the rate and speed and of the drone.

[0207] In the example of operation of the industrial system 30, the circuit board 10 is configured to handle any conflicts between gesture and voice control. For example, the voice command can always prevail, which, when given, overrides any gestural commands; this is particularly useful in the case of less experienced operators who can stop the hovering drone simply by pronouncing “stop”, independently of any gestural commands that may have been given in moments of excitement. Alternatively, for example, one can make one prevail between voice command and gesture command, based on criteria determined by the industrial system 30 itself by means of the Al accelerator 12, which implements Deep Learning algorithms for the recognition of the optimal conflict resolution strategy, based also on potential collisions with objects also detected by means of the drone's vision system.

[0208] In the example of operation of the industrial system 30, it is provided for an object recognition and automatic landing phase, in which: the industrial system 30 can be configured to recognise visual markers or specific objects for automatic landing; when the drone identifies the marker on the landing platform, it performs a controlled descent and lands safely. In this implementation, the Applicant provides using, as an Al accelerator 12, a Hailo-8™ or Hailo-H10™ processor from the company Hailo or an Ara-2™ processor from the company Kinara, suitable for the implementation of Deep Learning algorithms for the recognition of hand gestures and their position (at a speed of at least 30 frames per second), while for the CPU 14 it provides using a processor of the LMX8M™ Plus or LMX95 family from the company NXP for operator recognition, drone control and to delegate Deep Learning operations to the Al accelerator 12 processor.

[0209] As a further example, the Applicant implemented an industrial system 30 to provide object identification and inventory management, for example applicable to the automatic monitoring of the contents of a refrigerator. In this example, the industrial system 30 is able to identify the food in the refrigerator in real time and manage a digital inventory without the need for manual input by a user.

[0210] In this example of operation of the industrial system 30, it is provided for an image acquisition and object recognition phase, in which: a camera inside the refrigerator captures images of the contents of the refrigerator, in particular every time the door is opened and the refrigerator light is switched on; the industrial system 30 processes the images in real time using computer vision and Deep Learning algorithms to recognise the food items present (for example bottles, fruit, vegetables, dairy products, food packaging); the identified objects are automatically catalogued and the information stored in a database.

[0211] In the example of operation of the industrial system 30, it is provided for an inventory update phase, in which: each time new food is added or removed, the industrial system 30 updates the inventory according to the changes detected in the images; the user can access the inventory by means of a screen built into the refrigerator or by means of an application connected to the refrigerator; an icon or visual label can indicate whether a food has just been added, removed or is still present.

[0212] In the example of operation of the industrial system 30, it is provided for a notification phase on food availability, in which: if a certain food is no longer present in the refrigerator, the industrial system 30 can send a notification to indicate its absence, by means of an output device such as a module connected to the Internet; the user can receive a summary of the contents at any time to know what is available without having to open the refrigerator, for example by accessing a dedicated application on his / her smartphone. In this implementation, the Applicant provides using, as an Al accelerator 12, a Hailo-8™ processor from the company Hailo suitable for the implementation of Deep Learning algorithms for food recognition, while for CPU 14 he provides using a processor of the LMX8M™ Plus family from the company NXP for image acquisition and for sending notifications, as well as to delegate Deep Learning operations to the Hailo-8™processor.

[0213] As a further example, the Applicant implemented an Industrial System 30 for a combined process of semantic segmentation and real-time object detection; this process allows complex scenes to be analysed and objects to be identified accurately, for advanced machine vision applications.

[0214] In this example of operation of the industrial system 20, it is provided for an object recognition and localisation process, for example a detection and classification of objects within an image or video stream.

[0215] In the example of operation of the industrial system 30, it is provided for an image acquisition and pre-processing phase in which: a camera acquires an image or video frame; CPU 14 resizes and normalises the image to make it compatible with the selected Al model; the CPU 14 applies contrast and brightness correction to optimise recognition.

[0216] In the example of operation of the industrial system 30, it is provided for an Al inference phase for object recognition, in which: the image is further processed by an optimised model for object detection; the Al accelerator 12 performs low- latency inference, identifying the type and position of objects in the image; the detected objects are represented with bounding boxes, each associated with a category and a confidence score.

[0217] In the example of operation of the industrial system 30, it is provided for a data output and use phase, where: the results are sent to a user interface or control system to trigger automatic responses; in applications such as industrial monitoring or robot control, the information provided by the industrial system 30 can be used to identify anomalies or guide operational decisions.

[0218] In this example of operation of the industrial system 20, it is provided for a semantic segmentation process for the identification of each pixel of the image or video frame, in which an image is divided into labelled areas, pixel by pixel, distinguishing objects and backgrounds with greater precision. In the example of operation of the industrial system 30, it is provided for an image pre-processing and normalisation phase, in which: the image is acquired and pre- processed by the CPU 14 to fit the segmentation pattern;CPU 14 applies colour balancing and contrast enhancement operations to optimise recognition.

[0219] In the example of operation of the industrial system 30, it is provided for an Al inference phase for segmentation, in which: a network model processes the image; the Al accelerator 12 performs a real-time segmentation, assigning each pixel a label corresponding to the relevant class (for example: road, vehicle, person, building, etc.); a colour segmentation map is generated, highlighting the different categories in the image.

[0220] In the example of operation of the industrial system 30, it is provided for an output and application phase, in which: the result is displayed by a user on a screen or is used to make automated decisions; in industrial production applications, the segmentation provided by the industrial system 30 can, for example, be used to identify defects in a production line; in autonomous industrial systems, such as drones or self-driving vehicles, the segmentation provided by the industrial system 30 allows the environment to be mapped and, for example, to distinguish obstacles from passable areas.

[0221] In this implementation, the Applicant envisages using, as the Al accelerator 12, a Hailo-8™ or Hailo-H10™ processor of the company Hailo or an Ara-2™ processor of the company Kinara, suitable for the implementation of Deep Learning algorithms for the segmentation, while for the CPU 14 it envisages using a processor of the LMX8M™ Plus or LMX95™ family from NXP for image preprocessing and to delegate Deep Learning operations to the processor of the Al accelerator 12.

[0222] As will be clear from the foregoing description, in the device 1 according to the invention the "smart" part for executing artificial intelligence algorithms is delegated to dedicated electronic components (suitably the CPU 14 and the Al accelerator 12 arranged on the electronic board 10) while the interface part to external input / output peripherals is mainly delegated to the carrier board 20 whose main function is to interface with the dedicated electronic components and the input / output peripherals 31 . In addition, the “smart” part is configured to be mounted on the interface part in a removable manner. This, as described above, makes it possible to provide a flexible solution and to realise reconfigurable industrial systems. For example, if it is necessary to add new peripherals 31 or to replace peripherals 31 with others having different connectors than the connectors 21 present on the interface carrier board 20, it will be possible to remove circuit board 10 from the current carrier board 20 in order to mount it on a new carrier board 20 equipped with the required new connectors 21 . In addition, in the event that it is necessary to replace the Al accelerator 12 and / or CPU 14 with others having different characteristics (for example having greater computing capability) than those in the current electronic board 10, it will be possible to remove the current electronic board 10 from the carrier board 20 to mount on the same carrier board 20 a new electronic board 10 equipped with the new Al accelerator 12 and / or the new CPU 14.

Claims

C L A I M S1. Industrial system (30) comprising:- input / output peripherals (31 ) configured for the implementation of a specific industrial application;- an interface carrier board (20) comprising data exchange connectors (21 ) for the connection to said input / output peripherals (31 );- an electronic board (10) comprising a CPU (14) and a dedicated artificial intelligence accelerator (12), distinct from the CPU (14), the CPU (14) and the artificial intelligence accelerator (12) being connected to each other by means of data exchange interfaces; wherein:- at least one software package suitably configured for the implementation of said specific industrial application is installed on the artificial intelligence accelerator (12) and on the CPU (14);- the electronic board (10) is removably mounted on the carrier board by means of mechanical connection members (1 1 );- the electronic board (10) and the carrier board (20) are connected to each other by means of data exchange interfaces (13, 23);- the interface carrier board (20) is connected to said input / output peripherals (31 ) by means of said data exchange connectors (21 ).

2. Industrial system (30) according to claim 1 , wherein the electronic board (10) has a compact form factor with sizes less than 60 mm x 50 mm x 5 mm.

3. Industrial system (30) according to any one of the preceding claims, wherein the carrier board (20) is devoid of electronic components having computing capability suitable for the execution of artificial intelligence algorithms.

4. Industrial system (30) according to any one of the preceding claims, wherein the electronic board (10) also comprises a GPU distinct from the artificial intelligence accelerator (12).

5. Industrial system (30) according to any one of the preceding claims, wherein the CPU (14) comprises a further artificial intelligence accelerator integrated therein.

6. Industrial system (30) according to any one of the preceding claims, wherein the data exchange connectors (21 ) comprise at least one connector selected from the group comprising: HDMI connector, USB connector, Ethernet connector,serial connectors (for example of the RS-232 and RS-485 type), GPI connectors, GPO connectors, I2C connectors, SPI connectors, PCI connectors, PCIe connectors, connectors for memory boards, connector for external fan, Ml PI-DSI connectors, MIPI-CSI connectors, SDIO connectors, PCM connectors, PWM connectors, Canbus connectors.

7. Industrial system (30) according to any one of the preceding claims, wherein the input / output peripheral (31 ) is selected from the group comprising: a sensor, a control device, a transducer, a display, a processor, an industrial robot, a memory medium, a fieldbus.

8. Industrial system (30) according to any one of the preceding claims, wherein the industrial application is related to at least one sector selected from the group comprising the automotive sector, the smart homes and / or cities sector, the internet of things sector, the industrial production plant sector, the automation sector, the robotics sector, the drone sector, the quality control sector, the 3D vision systems sector, the stereovision sector, the smart camera sector.

9. Industrial system (30) according to any one of the preceding claims, wherein the industrial application is related to at least one application selected from the group comprising: recognition and / or detection of people and / or objects, recognition of the position of people and / or objects, 3D pose estimation, 3D depth estimation, gesture recognition, command of industrial robots, detection of anomalies of operation of a device or plant, control of operation of a device or plant, detection of anomalies in a product, surveillance operations, monitoring operations, assisted and automated driving operations of vehicles and the like.

10. Reconfigurable industrial system (30) comprising:- input / output peripherals (31 ) configured for the implementation of a specific industrial application;- an interface carrier board (20) comprising data exchange connectors (21 ) for the connection to said input / output peripherals (31 );- electronic components (12, 14) used for the execution of artificial intelligence algorithms suitably configured for the implementation of said specific industrial application; wherein:- said electronic components (12, 14) are arranged on an electronic board (10) which is removably mounted on the carrier board (20) by means of mechanical connection members (1 1 );- the electronic board (10) and the interface carrier board (20) are connected to each other by means of data exchange interfaces (13, 23);- the interface carrier board (20) is connected to said input / output peripherals (31 ) by means of said data exchange connectors (21 ).11 . Method for realizing an industrial system (30) comprising:- providing input / output peripherals (31 ) configured for the implementation of a specific industrial application;- providing an interface carrier board (20) comprising data exchange connectors (21 ) for the connection to said input / output peripherals (31 );- providing an electronic board (10) comprising a CPU (14) and a dedicated artificial intelligence accelerator (12), distinct from the CPU (14), wherein the CPU (14) and the artificial intelligence accelerator (12) are connected to each other by means of data exchange interfaces;- providing in a server (40) a collection of software comprising a plurality of software packages, each software package comprising a series of software programs that are jointly configured for the installation on the artificial intelligence accelerator (12) and on the CPU (14) and for the implementation of a respective industrial application among a plurality of industrial applications;- obtaining from the server (40) at least one software package, among said plurality of software packages, suitably configured for the implementation of said specific industrial application among said plurality of industrial applications and installing said at least one software package thus obtained on the artificial intelligence accelerator (12) and on the CPU (14);- removably mounting the electronic board (10) on the carrier board (20), the electronic board (10) and the carrier board (20) being connected to each other by means of data exchange interfaces (13, 23);- connecting the interface carrier board (20) to said input / output peripherals (31 ) by means of said data exchange connectors (21 ).

12. Method for controlling the operation of an industrial system (30) comprising:- providing input / output peripherals (31 ) configured for the implementation of a specific industrial application;- providing an interface carrier board (20) comprising data exchange connectors (21 ) for the connection to said input / output peripherals (31 );- providing an electronic board (10) comprising a CPU (14) and a dedicated artificial intelligence accelerator (12), distinct from the CPU (14), wherein the CPU (14) and the artificial intelligence accelerator (12) are connected to each other bymeans of data exchange interfaces;- obtaining from a server (40) at least one software package suitably configured for the implementation of said specific industrial application and installing said at least one software package thus obtained on the artificial intelligence accelerator (12) and on the CPU (14);- removably mounting the electronic board (10) on the carrier board (20), the electronic board (10) and the carrier board (20) being connected to each other by means of data exchange interfaces (13, 23);- connecting the interface carrier board (20) to said input / output peripherals (31 ) by means of said data exchange connectors (21 );- operating the input / output peripherals (31 ) by running said at least one software package installed on the artificial intelligence accelerator (12) and on the CPU (14).

13. Electronic device (1 ) comprising:- a carrier board (20) of interface to input / output peripherals (31 ) configured for the implementation of an industrial application, the carrier board (20) comprising data exchange connectors (21 ) for the connection to said input / output peripherals (31 );- an electronic board (10) comprising a CPU (14) and a dedicated artificial intelligence accelerator (12), distinct from the CPU (14), the CPU (14) and the artificial intelligence accelerator (12) being connected to each other by means of data exchange interfaces; wherein:- the electronic board (10) and the interface carrier board (20) are provided with data exchange interfaces and the electronic board (10) comprises mechanical connection members (1 1 ) configured to removably mount the electronic board (10) on the carrier board (20); and- the electronic board (10) is mounted on the carrier board (20) by means of said connection members (1 1 ) and interfaces with the carrier board (20) by means of said data exchange interfaces.

14. Kit for industrial systems comprising:- an electronic board (10) comprising a CPU (14) and a dedicated artificial intelligence accelerator (12), distinct from the CPU (14), the CPU (14) and the artificial intelligence accelerator (12) being connected to each other by means of data exchange interfaces;- a collection of software stored in a server (40) and comprising a plurality ofsoftware packages, each software package comprising a series of software programs that are jointly configured for the installation on the artificial intelligence accelerator (12) and on the CPU (14) and for the implementation of a respective industrial application among a plurality of industrial applications; wherein the electronic board (10) comprises:- mechanical connection members (1 1 ) configured to removably mount the electronic board (10) on a carrier board (20) of interface to input / output peripherals (31 ) that are configured for the implementation of a specific industrial application among said plurality of industrial applications; and - a data exchange interface between the electronic board (10) and the carrier board (20).

Images