Method for providing services using a chatbot in a factory and an associated system

Abstract

The invention relates to a method for providing services using a chatbot in a factory, the method being implemented by a computer system that executes programmed instructions stored in the memory of one or more processors of the computer system to construct at least one knowledge graph, characterized in that the method includes a step of carrying out a process for creating at least one knowledge graph. The invention also relates to a system that performs a method for providing services using a chatbot in a factory, in which the system includes a computer system that executes programmed instructions stored in the memory of one or more processors of the computer system to perform the method of the invention. The invention further relates to a tire manufacturing plant comprising the disclosed system. Figure for the abstract: Fig. 6

Description

Title of the invention: Method for providing services using a conversational robot in a factory and an associated system technical field

[0001] The invention relates to a method of providing services using a chatbot in a factory. An operator of an associated system who performs the method of the invention obtains a contextualized response to a request formulated in a vocabulary specific to them. Context

[0002] It is understood that an operator in a factory may find themselves in a situation where a problem or question arises. The operator is not always experienced and needs to formulate a query using vocabulary specific to their job and environment. Such a query must be formulated based on knowledge that is easily accessible to all operators, regardless of their different skill levels. The system used by the operator to access the knowledge base will typically operate in three stages: - identify what the operator is asking for (the "understanding" part); - search the knowledge base for the most appropriate answer (the "search" section); and - return the corresponding response to the operator (the "return" part).

[0003] The "understanding" part requires that the system possess a certain adaptability to understand the context in which the operator is acting and therefore to be able to interpret the specific vocabulary within that context. For example, consider the term "drum." In the tire manufacturing industry, a "drum" is a type of large metal cylinder around which semi-finished products are placed to manufacture a tire. However, in an orchestra, a "drum" is a type of large cylinder that is struck (for example, using hands and / or drumsticks) to make music.

[0004] The "research" aspect is a significant challenge in manufacturing companies today because the accumulated experience is stored in the form of documents designed by humans for humans. These references are unsuitable for automated searching. Therefore, this aspect requires specific methods to overcome these difficulties.

[0005] To achieve the "restitution" part, it is possible to exploit the capabilities of natural language processing (or "NLP"). NLP is a subfield of artificial intelligence that focuses on interactions between computers (synthesized) and human (natural) languages. In particular, NLP concerns how to program computers to process and analyze large amounts of data in natural language. Developing NLP applications is challenging because computers have traditionally required humans to "speak" to them in a precise, unambiguous, and highly structured programming language or through a limited number of clearly stated voice commands.NLP is the technology behind chatbots, intelligent virtual assistants (for understanding inflection and tone), online translation services, Google search results (which use NLP to proactively understand the searcher's intent and quickly provide relevant results), predictive text (auto-completion, auto-correction), and other digital tools (see Kaviraju, “Leveraging Capabilities of NLP in Manufacturing Industries,” KR Knowledge World (industry40.co.in) (October 2021)) (“the Kaviraju reference”). This technology requires a very large volume of data and significant computing power for training (the GPT-3 model, for example, relies on hundreds of billions of words for training and 175 gigabytes to store network parameters). The problem of comprehension remains difficult to solve (for example, the problem of the double meaning of the word “drum”).

[0006] In the prior art, there are solutions in which data preprocessing using NLP techniques can be applied to historical machine failure reports to extract summary data. For example, US patent 9,898,527 discloses a method for retrieving information performed by a computer device that analyzes a received request to identify a subject of interest. The computer device determines when the subject of interest identified in the received request corresponds to one or more ontological entities. In this case, the computer device identifies when the identified subject of interest is one of one or more existing subjects of interest.During the process, the computing device performs storage, associated with the identified subject of interest, of one or more filters comprising one or more new property conditions in the received query; one or more ontological property conditions associated with one of the mapped ontological entities; and one or more existing property conditions associated with one or more existing subjects of interest. The computing device can automatically construct an updated query based on one or more stored filters; and, on this basis, it can retrieve and provide one or more results based on the updated query.

[0007] In another example, publication EP3751503 discloses the use of a knowledge graph in a conversational robot (or "chatbot"). Various embodiments provide a device to support a chat service capable of providing simultaneously various services through a plurality of chatbots.

[0008] Despite these solutions, as explained in the Kaviraju reference, industries cannot hire only experienced operators. Therefore, there is always a mix of operators with varying skill levels. An operator is expected to handle a machine problem independently. On the one hand, experienced operators are likely to refer to historical machine failure reports before resolving the problem. On the other hand, less experienced operators must search for historical reports without knowledge of the applicable terminology. Consequently, it is necessary to adopt an approach that helps operators, whether highly experienced or less experienced, to review historical reports (being the documents available within the operator's predetermined role), understand them quickly, and repair the machine as quickly as possible.

[0009] The use of a knowledge graph makes it possible to overcome a lack of homogeneity in collective intelligence. In a well-known example, Google uses the structured database Wikidata as a knowledge graph to populate and supplement its search query responses. This database has been built over time by the global community. Management rules and tools have been developed in parallel to support its development (see https: / / www.wikidata.org / wikiAVikidata:Introduction / fr). Initiatives such as the SAREF ontology ("Smart Applications REFerence") (see https: / / saref.etsi.org / index.html) allow for advanced management similar to source code management and knowledge graphs. However, they do not allow for the automatic generation of a knowledge graph from text.

[0010] The disclosed invention takes advantage of this type of graph and the representation of the set of meanings that can be associated with a word. It is understood that the term "meaning" refers to the definition of a word in the context of its use. It is also understood that the term "concept" refers to the set of ideas that come to mind around a word in its context (for example, the ideas "sticks," "music," "semi-finished goods," and / or "pneumatics" in relation to the word "drum"). From a cognitive point of view (i.e., in the brain), concepts can be associated, in addition to words, with images, smells, sounds, memories and other triggers of understanding and meaning given to a text (see Dounia Lakhzoum, Semantic Organisation and Generalisation Mechanisms of Abstract Concepts, Psychology, Université Clermont Auvergne; Laboratoire de Psychologie Sociale et Cognitive (LAPSCO)(2021).The concept of learning and the link between concepts are still poorly understood, which explains the difficulty of implementing them in artificial intelligence and the need for human validation. knowledge model. Thus, it is on the generation of this knowledge graph that the disclosed invention applies NLP techniques to historical machine failure reports in order to extract summary data, regardless of the experience level of the operators involved. Summary of the invention

[0011] The invention relates to a method of providing services using a conversational robot in a factory, the method implemented by a computer system enabling the execution of programmed instructions stored in the memory of one or more processors of the computer system to construct at least one knowledge graph, characterized in that the method comprises a step of carrying out a process of creating at least one knowledge graph, the process comprising the following steps: - a first step consisting of developing a control vocabulary to encourage the use of the same word for a particular meaning; - a second step consisting of developing a thesaurus whose equivalence of terms used in a predetermined activity makes it possible to identify the terms which relate to a single concept of the predetermined activity; - a third step consisting of creating a taxonomy that defines a hierarchy with parent-child relationships; and - a final step of creating at least one knowledge graph to capture specific instances of the relationship; in which the processor(s) include an analysis application execution module which performs the processing of terms collected from documents available in the predetermined activity, of which the processor(s) are capable of executing programmed instructions stored in memory to carry out the steps of the process.

[0012] In some embodiments of the method of the invention, the step of creating at least one knowledge graph includes a step of determining the different links between the terms identified during the second step of the process.

[0013] In certain embodiments of the process of the invention, the second step of the process includes a determination step, by the computer system, where each term identified in a request received from an operator of the computer system corresponds to one or more predetermined ontological entities.

[0014] In certain embodiments of the process of the invention, during the last step of the process, the computer system retrieves and provides one or more knowledge graphs based on an automatically constructed updated query.

[0015] In certain embodiments of the process of the invention, the second step of the process comprises a step of decomposing the identified terms into words and their concepts, during which the documents entered into the computer system are weighed in a way that makes the most use of words and their associated concepts.

[0016] In some embodiments of the process of the invention, the last step of the process includes a step of building a model by statistical learning to create a natural language processing (NLP) model on the basis of the documents entered into the computer system, so that the constructed model contains proximity values ​​between the identified terms in order to extract the words associated with each other.

[0017] In certain embodiments of the process of the invention, the last step of the process further includes a step of adding one or more knowledge graphs created beforehand.

[0018] In certain embodiments of the process of the invention, wherein the predetermined activity includes the manufacture of tires.

[0019] The invention also relates to a system that performs a method of providing services using a chatbot in a factory, in which the system comprises: - a computer system enabling the execution of programmed instructions stored in the memory of one or more processors of the computer system to carry out the disclosed process;

[0020] - a knowledge base to which one or more knowledge graphs are linked to a predetermined activity; - at least one communication network incorporating one or more communication servers; and - at least one user interface communicating with the processor(s) of the computer system.

[0021] In certain embodiments of the system of the invention, the knowledge base includes documents available in the predetermined activity.

[0022] The invention further relates to a factory for the manufacture of tires, comprising the disclosed system.

[0023] Other aspects of the invention will become evident from the following detailed description. Brief description of the drawings

[0024] The nature and various advantages of the invention will become more evident upon reading the following detailed description, together with the accompanying drawings, in which the same reference numbers designate identical parts throughout, and in which: [Fig. 1] Figure [Fig. 1] represents one way of implementing a knowledge graph built in the field of tire manufacturing. [Fig.2] Fig.2 represents an example of terms used in the construction of a knowledge graph in the field of tire manufacturing. [Fig.3] [Fig.3] represents the discriminating terms identified among the terms represented in [Fig.2]. [Fig.4] Figures 4 and 5 represent the relative distances between the terms used in the construction of a knowledge graph in the field of tire manufacturing. Detailed description

[0025] The invention relates to a method for providing services using a chatbot in a factory (or "method"). The method of the invention is implemented by a computer system (or "system") that executes programmed instructions stored in the memory of one or more processors of the system to construct at least one knowledge graph. The method of the invention includes a process for creating one or more knowledge graphs (or "process") implemented by the system to accommodate an unstructured database (i.e., not formatted to facilitate question / answer interactions). The system requires the following elements: - A knowledge base (including documents available in a predetermined activity); - One or more knowledge graphs linked to a predetermined activity, the graphs of which have been created beforehand based on the words used in the daily work of a predetermined activity; - At least one user interface; and - A computer structure adapted to work with algorithms called "statistics".

[0026] Referring now to the figures, in which the same numbers identify identical elements, [Fig. 1] represents an embodiment of a knowledge graph constructed in the field of tire manufacturing. A knowledge graph allows for the modeling of knowledge. This model is based on concepts: ideas that are generally identified by a word or expression and that are linked together by a proximity relationship. It is understood that such a knowledge graph could be constructed with respect to any field.

[0027] As shown in [Fig. 1], a BNS shaping machine is depicted which allows the transformation from a cylindrical shape to a toroidal shape, which will be that of a tire. The BNS machine consists of a DRUM, and its purpose is to produce tire peaks on a CARCASS (for example, in rap (bringing the beads closer together while subjecting the central part to pressure that will stretch and bulge the rubber layers). On the CARCASS, a PRODUCT LAYER is performed in which the top elements are added (including, but not limited to, the reinforcement plies, the top plies, the smooth profile that will become the tread, and the various edge rubbers). Each PRODUCT LAYER must meet a predefined QUALITY standard. During a production session, the mechanics of the BNS machine may malfunction, and a component such as the DRUM has a probability of FAILURE. If there is a FAILURE, there is a PROBLEM (for example, a FAILURE can affect a PRODUCT LAYER, which threatens QUALITY).

[0028] Referring again to [Fig.1], by way of example, the different links are weighted (0.2 between BNS and CARCASS or between BNS and DRUM), which means that, in an industrial context of tire manufacturing, people associate the BNS with the DRUM as much as with the CARCASS.

[0029] When a question is posed to the system, it breaks down the sentence into words or phrases and searches the knowledge graph for concepts of ranks 1, 2, ..., N (and their weights) associated with them. This set of words and their weighted concepts is then used for a simple search of available documents, which are assigned a weight based on the number of occurrences of each word and concept. The results are then sorted by weighting the documents, with the most relevant being those that have made the most frequent use of the words and their associated concepts. Thus, a document that refers to all the words in the question will have a high weight. A document that refers to only one concept unrelated to one of the words will, conversely, have a significantly lower weight.

[0030] The robustness of the system of the invention is therefore largely linked to the accuracy of the knowledge graphs. In the disclosed invention, the addition of simple graphs improves search performance without increasing design cost or system complexity. The system is thus described as a knowledge graph-based system.

[0031] As an example, let's take as input the question "The BNS has a PROBLEM, what to do?" Referring to [Fig. 2], if we use only the words BNS and PROBLEM (underlined in [Fig. 2]), Document 1 appears first and Document 4 is not even present in the results. If we use related terms from the knowledge graph (for example, the terms CARCASS, PRODUCT INSTALLATION, BREAKDOWN, and QUALITY underlined in [Fig. 2]), Documents 2 and 3 appear before Document 1, and Document 4 enters the results, which is justified in the context of a tire manufacturing workshop.

[0032] By further referring to [Fig. 3], the discriminating terms between Documents 2 and 3 in [Fig. 2] are "BREAKDOWN" and "QUALITY," which allow for potentially very different responses, focusing on either breaking down or quality issues. Starting with the word PROBLEM, it's impossible to distinguish between a quality problem and a breakdown: the distances PROBLEM-BREAKDOWN and PROBLEM-QUALITY are identical.

[0033] Referring further to Figures 4 and 5, starting from the word BNS, the BNS-FAILURE distance is shorter than the BNS-QUALITY distance if the term DRUM is included in the graph ([Fig. 5]) than if DRUM is not ([Fig. 4]). In each case, the shortest path will make the difference and will direct the user towards documents oriented towards "FAILURE," intended for maintenance, or "QUALITY," for the development department. Thus, the task is to find the most efficient but least complex graph, because complexity will hinder the management and maintenance of the graphs. They can be more or less detailed, but also oriented towards a predetermined sector of activity (here, quality or maintenance) to address problems in different professions.

[0034] It is understood that generating knowledge graphs of the invention by humans is extremely time-consuming and requires significant professional skills as well as mastery of knowledge management tools. A dictionary-based methodology for a predetermined activity involves analysis using statistical algorithms employed in NLP to directly obtain the concepts associated with a word. Thus, automated processing based on NLP accelerates the creation of knowledge graphs.

[0035] The knowledge graph construction process relies on ontologies, that is, sets of concepts and categories within a domain or profession, which demonstrate their properties and the relationships between them. By default, NLP will operate using statistical learning to create an NLP model based on the document database, which includes the documents available in the predetermined activity (for example, tire manufacturing). This model contains proximity values ​​between words to extract words that are "regularly" associated with one another, thus creating a weighted knowledge graph that will serve as a basis.

[0036] The method of the invention includes a step of implementing a process for creating at least one knowledge graph (or "process"). The process includes a first step consisting of developing a control vocabulary to encourage the use of the same word for a particular meaning (for example, associating the term "drum" to refer to the large metal cylinder used in tire manufacturing and not to the large musical cylinder). The next step of the process consists of developing a thesaurus whose term equivalences allow The process begins by identifying terms that relate to a single concept within a predetermined activity (for example, in cases where the predetermined activity concerns tire manufacturing, the terms "elastomeric" and "synthetic" refer to the concept of a rubber compound, but these terms are not synonymous). The next step involves creating a taxonomy that defines a hierarchy with parent-child relationships (for example, one element being classified as part of another, such as a drum being part of a BNS machine). Finally, the process includes a last step of creating a knowledge graph to capture specific instances of the relationship, while the ontology is generic. In the chosen predetermined activity (such as tire manufacturing), the quality of this graph depends on the volume of available documentation.

[0037] The data used (for example, documents 1 to 4 in [Fig. 2]) are optimized using machine learning algorithms (including, but not limited to, Naive Bayes, KNN, logical regression, SVM, and decision trees) and / or deep learning algorithms (including, but not limited to, encoders / decoders using LSTM / RNN, BERT, etc.). Furthermore, tools such as SAREF allow for graph merging by detecting inconsistencies, which are then resolved by a human. This enables an expert employee to improve a low-quality graph (resulting from NLP processing of the documents) by merging it with one from the dictionary or a public graph dealing with a known domain. This results in a high-quality graph with a substantial time saving.

[0038] With reference to [Fig. 6], an example of a knowledge graph is shown, the contents of which are found in a dictionary used in tire manufacturing. In this dictionary, the term "Scraper" is associated with a "Blade" applied against a "Cylinder" to bring the unused "Tumor" back to the "Bead." In this dictionary, the term "Bead" is associated with a "Quantity of Mixture" found on the rotating "Cylinders" of a finishing homogenizer (or "HF"), which is necessary and sufficient to obtain a homogeneous and viscoelastic "Mixture." On a "Calendar," the bead, by ensuring a regular "Feed," allows for a continuous sheet of mixture.

[0039] In this example, it is clear that the "contextualized" meanings of words are directly reusable and search efficiency is improved. The validation and management of knowledge graphs could rely on tools like SAREF to minimize their cost. The system of the invention therefore creates semi-automated, rather than entirely manual, knowledge graphs.

[0040] As used herein, “operator” refers to a single operator or a group of operators. An operator includes, but is not limited to, an individual operator of a A predetermined activity, an individual member of a team or group participating in a predetermined activity, one or more machines associated with an individual or team participating in at least one task of a predetermined activity, a digital community associated with a predetermined activity, and combinations and equivalents thereof. The operator may be a spectator attending a task, either physically or digitally (for example, by remotely managing a live predetermined activity to observe it in real time). As used here, “operator” may also refer to one or more others analyzing data collected in another environment. As used here, “operator” may also refer to any electronic system or device configured to receive command input and configured to automatically send data to at least one other operator.

[0041] As used herein, the term “method” or “process” may include one or more steps performed by at least one computer system comprising one or more processors to execute instructions that carry out the steps (including the construction of knowledge graphs). Any sequence of steps is given by way of example and does not limit the described methods to any particular sequence unless otherwise indicated.

[0042] The computer system that implements the method of the invention comprises a communication network incorporating one or more communication servers (or "servers") that manage incoming data to the system from various sources (for example, documents available in the predetermined activity). The communication network may include wired or wireless connections and may implement any data transfer protocol known to a person skilled in the art. Examples of wireless connections may include, without limitation, radio frequency (RF), satellites, mobile phones (analog or digital), Bluetooth®, Wi-Fi, infrared, ZigBee, local area network (LAN), wireless local area network (WLAN), wide area network (WAN), near field communication (NFC), other wireless communication configurations and standards, their equivalents, and combinations thereof.

[0043] The term "processor" (or, alternatively, the term "programmable logic circuit") refers to one or more devices capable of processing and analyzing data and comprising one or more software programs for processing such data (for example, one or more integrated circuits known to those skilled in the art as being included in a computer, one or more controllers, one or more microcontrollers, one or more microcomputers, one or more programmable logic controllers (or "PLCs"), one or more application-specific integrated circuits, one or more neural networks, and / or one or more other known equivalent programmable circuits). The processor includes software for processing the data captured by the elements associated with the computer system (and the corresponding data obtained) as well as software for identifying and locating variances and identifying their sources in order to correct them.

[0044] The processor(s) are operationally connected to a memory configured to store an application for analyzing terms representative of the predetermined activity (for example, tire manufacturing in a factory including the system). The processor(s) include an application execution module that processes the terms collected from available documents, and the processor(s) are capable of executing programmed instructions stored in memory to carry out the steps of the process.

[0045] Memory may include both volatile and non-volatile memory devices. Non-volatile memory may include solid-state memories, such as NAND flash memory, keep-alive memory (or KAM) for saving various operating variables while the processor is powered off, magnetic and optical storage media, or any other suitable data storage device that retains data when the system is powered off or loses its power supply. Volatile memory may include static and dynamic RAM that stores program instructions and data, including a learning application.

[0046] The system that implements the method of the invention comprises at least one user interface communicating with the processor(s) of the computer system. The user interface may incorporate one or more types of interfaces, including, but not limited to, a graphical user interface (or "GUI"), a command-line interface (or "CLI"), a menu-driven interface (or "MDI"), a form-based interface (or "FBI"), and / or a natural language user interface (or "NLI"). In one embodiment of the user interface, the user interface comprises a graphical user interface that communicates to the operator the terms enabling the creation of one or more knowledge graphs based on a query entered by the operator. On this basis, the computer system retrieves and provides one or more knowledge graphs based on an automatically constructed, updated query.

[0047] The system of the invention may include pre-programming of knowledge graph construction information. For example, an operating setting of a BNS machine may be associated with the parameters of the typical physical environment in which the BNS machine operates.

[0048] In embodiments of the invention, the system (and / or a plant comprising the system) can receive audio commands (including voice commands) or other audio data representing, for example, a step in or a pause in the construction of at least one knowledge graph from the available documents. A request to the system may include a request for the current state of an ongoing knowledge graph construction process. A generated response may be represented audibly, visually, tactilely (e.g., using a haptic interface), and / or virtually and / or augmented. This response, together with the corresponding data, may be stored in a neural network.

[0049] For all embodiments of the system, a monitoring system could be implemented. At least part of the monitoring system can be provided in a portable device such as a mobile network device (e.g., a mobile phone, a laptop computer, one or more portable network-connected devices (including augmented reality and / or virtual reality devices), network-connected wearable clothing, and / or any combination thereof). It is conceivable that detection and comparison steps could be performed iteratively.

[0050] The terms "at least one" and "one or more" are used interchangeably. The ranges presented as being "between a and b" encompass the values ​​"a" and "b".

[0051] Although particular embodiments of the disclosed apparatus have been illustrated and described, it will be understood that various changes, additions, and modifications can be made without departing from the spirit or scope of this disclosure. Therefore, no limitations should be imposed on the scope of the invention described except those set forth in the appended claims.

Claims

Claims

1. A system that performs a method of providing services using a chatbot in a factory, wherein the system comprises: - a computer system for executing programmed instructions stored in a memory of one or more processors of the computer system to perform a method comprising a step of performing a process of creating at least one knowledge graph, the process comprising the following steps: - a first step of developing a control vocabulary to encourage the use of the same word for a particular meaning; - a second step of developing a thesaurus whose equivalence of terms used in a predetermined activity makes it possible to identify terms that relate to a unique concept of the predetermined activity; - a third step of producing a taxonomy that defines a hierarchy with parent-child relationships;and - a final step of creating at least one knowledge graph to capture specific instances of the relationship; wherein the processor(s) comprise an analysis application execution module that performs the processing of the terms collected from the documents available in the predetermined activity, the processor(s) of which are capable of executing programmed instructions stored in the memory to perform the steps of the method; - a knowledge base on which one or more knowledge graphs are linked to a predetermined activity; - at least one communication network incorporating one or more communication servers; and - at least one user interface in communication with the processor(s) of the computer system.;

2. The system of claim 1, wherein the knowledge base comprises documents available in the predetermined activity.

3. The system of claim 1 or claim 2, wherein, during the second step of the process, the computer system determines different links between the identified terms.

4. The system of claim 3, wherein, during the second step of the process, the computer system identified each term in a query received from a computer system operator and determines a match to one of one or more predetermined ontological entities.

5. The system of claim 3 or claim 4, wherein, during the last step of the process, the computer system retrieves and provides one or more knowledge graphs based on an automatically constructed updated query.

6. The system of claim 5, wherein, during the second step of the process, the computer system decomposes identified terms into words and their concepts, such that documents input to the computer system are weighed in a manner that most makes use of the words and their associated concepts.

7. The system of claim 6, wherein, during the last step of the process, the computer system builds a model by statistical learning to create a natural language processing (NLP) model based on the documents input to the computer system, such that the built model contains proximity values ​​between the identified terms so as to extract words associated with each other.

8. The system of any one of claims 1 to 7, wherein, during the last step of the process, the computer system adds one or more previously created knowledge graphs.

9. The system of any one of claims 1 to 8, wherein the predetermined activity comprises manufacturing the tires.

10. A plant for manufacturing tires, comprising the system of any one of claims 1 to 10.

Images