Tool wear evaluation and / or tool wear prediction network

Abstract

A tool wear evaluation and / or tool wear prediction network (10) is proposed, comprising: a plurality of optical tool measuring devices (12, 14, 16, 18) which are provided at least for measuring tools (26) and are located at at least two geographically remote locations, a plurality of tool usage installations (20, 22, 24), such as machine tools, CNC machining centres, robotic installations, tool management systems, etc. which use the measured tools (26), wherein at least one of the optical tool measuring devices (12, 14, 16, 18) is geographically co-located with one or more of the tool usage installations (20, 22, 24), and a centrally localised or delocalised evaluation and / or prediction computer system (28), wherein at least the optical tool measuring devices (12, 14, 16, 18) and the tool usage installations (20, 22, 24) are directly or indirectly connected to the evaluation and / or prediction computer system (28) in a data-transmitting manner, wherein the evaluation and / or prediction computer system (28) comprises at least one machine learning algorithm which has been specially trained and can be further trained for a wear analysis of tools (26), in particular of tool blades.

Description

[0001] ZO 17913 WO

[0002] October 21, 2025

[0003] Tool wear evaluation and / or tool wear prediction network

[0004] State of the art

[0005] The invention relates to a tool wear evaluation and / or tool wear prediction network according to claim 1, a method for training a machine learning algorithm specifically trained and further trainable for tool wear analysis according to claim 11, a method for evaluating and / or predicting tool wear according to claim 12, and a centrally localized or delocalized evaluation and / or prediction computer system according to claim 13.

[0006] Tools are already being inspected for wear before being used in tooling applications. These inspections are carried out manually or with computer assistance. The results and quality of these inspections can vary considerably and often depend significantly on the experience and / or subjective assessment of the personnel involved and / or the fine-tuning of the computer programs used, as well as on the scope of comparative and / or reference wear characteristics available to the person or the computer program.

[0007] The object of the invention is, in particular, to provide a generic device with advantageous properties regarding the quality and significance of wear assessments of tools. This object is achieved according to the invention by the features of claim 1, while advantageous embodiments and further developments of the invention can be found in the dependent claims. ZO 17913 WO

[0008] Advantages of the invention

[0009] A tool wear evaluation and / or tool wear prediction network is described, comprising: a plurality of optical tool measuring devices intended for measuring tools at least once, which are located at at least two geographically distant locations; a plurality of tool application systems, such as machine tools, CNC machining centers, robotic systems, tool management systems, etc., which apply measured tools, wherein at least one of the optical tool measuring devices is geographically co-located with one or more of the tool application systems, and wherein preferably at least one of the optical tool measuring devices is located at a location far removed from each of the tool application systems; and a centrally located or delocalized evaluation and / or prediction computer system.wherein at least the optical tool measuring devices and the tool application systems are networked with the evaluation and / or prediction computer system directly or indirectly, e.g. via the optical tool measuring device co-located with one of the tool application systems, wherein the evaluation and / or prediction computer system comprises a machine learning algorithm specifically trained and further trainable for wear analysis of tools, in particular tool cutting edges, wherein the evaluation and / or prediction computer system has a training interface which is intended to receive training data in the form of optical measurement data, in particular images, of at least partially worn tools in combination with one or more wear information, for example, number of machining hours of the tools, previous machining tasks of the tools,to receive previously processed workpieces / materials, machining parameters of the tool application systems when using the tools, for further training of the machine learning algorithm from the optical tool measuring devices, and wherein the evaluation and / or prediction computer system is an evaluation and / or ZO 17913 WO,

[0010] A predictive interface is proposed, designed to receive optical measurement data, in particular the optical measurement data of a tool used or to be used in one of the tool application systems, from one of the optical tool measuring devices geographically colocated with this tool application system, and, following at least one wear analysis performed by the specially trained and further trainable machine learning algorithm, to send an evaluation of the tool wear and / or a prediction of the tool wear, in particular a remaining tool life, back to the (same) tool application system and / or to the (same) optical tool measuring device. This advantageously allows for high-quality and informative tool wear assessments.Advantageously, wear information from many different locations and application areas of tools can be aggregated, enabling optimized and / or particularly precise wear predictions and / or analyses. Advantageously, a particularly powerful AI can be generated and used for future analyses. Advantageously, the prediction and / or analysis results continuously improve and can adapt quickly and flexibly to changes (e.g., in tool geometries or workpiece machining techniques). Advantageously, high-quality and / or state-of-the-art wear assessments and / or tool life predictions can be made easily and user-friendly available to a large group of tool users.Advantageously, an objective wear assessment, which is particularly independent of individual experience, can be made easily and user-friendly available to a large group of tool users. Thanks to the tool wear evaluation and / or tool wear prediction network, tools and / or tool application systems can be operated particularly efficiently, and especially cost-effectively. ZO 17913 WO.

[0011] The tool wear evaluation and / or tool wear prediction network is, in particular, a data transmission-related connection of several electronic devices, especially their computing and / or control units, which has at least the purpose of sharing data and / or (computing) resources. The electronic devices participating in / connected to the tool wear evaluation and / or tool wear prediction network are physically (e.g., via cable) and / or wirelessly (e.g., via WLAN or other short-range and / or long-range radio protocols) connected to each other. An "optical tool measuring device" is understood to mean, in particular, a device that is designed to optically detect at least one length, at least one angle, at least one contour, and / or at least one external shape of a tool, at least partially, preferably completely.The optical tool measuring device is preferably based on camera-based and / or image analysis-based optical measuring methods for tools, such as transmitted light measurement, in particular transmitted light contour measurement, or preferably reflected light measurement, in particular reflected light surface measurement, of tools. One or more of the optical tool measuring devices can be optical tool presetting and tool measuring devices such as those distributed by E. ZOLLER GmbH & Co. KG Setting and Measuring Devices (Pleidelsheim, Germany, www.zoller.info) at the time of preparation and submission of this document. The tools are designed in particular as shank tools, preferably as rotary shank tools, for example, drills, milling cutters, profile tools, and / or reamers. Other tools, e.g., shankless tools or complete tools, etc., are also possible.However, they can also be processed by the tool wear evaluation and / or tool wear prediction network. The tool measuring devices participating in the tool wear evaluation and / or tool wear prediction network are preferably arranged, at least partially, across ownership, company boundaries, and / or countries. ZO 17913 WO.

[0012] However, it is also conceivable that the tool wear evaluation and / or tool wear prediction network only includes facilities belonging to a single owner / company or country. For example, a large corporation could simply connect its various plants to the tool wear evaluation and / or tool wear prediction network. The tool application systems preferably apply the measured tools to workpieces or the like and / or organize tools within a facility, e.g., a plant.

[0013] Geographically colocated objects / systems / devices are preferably located in the same geographical area, e.g., on the same factory premises, in the same factory hall, or preferably in the same room of the same building. Preferably, geographically colocated objects / systems / devices are located in geographical proximity to one another. Preferably, the time required to transfer a tool between two colocated objects / systems / devices at walking speed is less than 10 minutes, more preferably less than 5 minutes. Preferably, the shortest distance between two colocated objects / systems / devices is less than 1000 m, more preferably less than 500 m, and most preferably less than 100 m.An optical tool measuring device geographically colocated with a tool application system is specifically designed to optically measure tools used in the associated tool application system immediately before and / or immediately after use, particularly for detecting wear characteristics. The colocated optical tool measuring device is preferably permanently assigned to the respective tool application system within the tool wear evaluation and / or tool wear prediction network.

[0014] A centrally located evaluation and / or prediction computer system can, for example, be implemented by a server or a server cluster. A delocalized evaluation and / or prediction computer system can, for example, be implemented by a server network distributed across various geographical locations or by a cloud computing infrastructure. In particular, the delocalized evaluation and / or prediction computer system can be a cloud-based evaluation and / or prediction computer system. Preferably, the optical tool measuring devices and / or the tool application systems of the tool wear evaluation and / or tool wear prediction network are each wirelessly and / or via a wired connection to the evaluation and / or prediction computer system.Alternatively or additionally to a direct data connection between one of the tool application systems and the evaluation and / or prediction computer system, an indirect data connection via an intermediary device, e.g., an optical tool measuring device colocated with one of the tool application systems, is also conceivable. In the case of an indirect data connection, it is advantageous that no software needs to be installed on the tool application system / machine tool itself.Instead, the optical tool measuring device, which is connected to the tool application system / machine tool and is already configured, can be designed to exchange tool data with the tool application system / machine tool, and preferably also to perform aggregation (linking tool image acquisition with usage data from the tool application system / machine tool for the imaged tool) and / or subsequent data transmission. A direct data transmission connection between several of the optical tool measuring devices, between several of the tool application systems, and / or between optical tool measuring devices and tool application systems is conceivable, but irrelevant for the function of the tool wear evaluation and / or tool wear prediction network.Preferably, at least some of the optical tool measuring devices and / or the tool application systems are free from a direct data transmission connection to another tool measuring device and / or another tool ZO 17913 WO.

[0015] Application system of the tool wear evaluation and / or tool wear prediction network.

[0016] The evaluation and / or prediction computer system can comprise one or more interacting, specially trained and further trainable machine learning algorithms. The machine learning algorithm(s) used for tool wear analysis (specifically trained and further trainable for this purpose) is / are preferably specifically trained to determine wear categories or wear assessments based on optical tool measurement data from optical tool measuring devices, such as reflected and / or transmitted light images. The trained machine learning algorithm(s) used for this purpose receive the optical tool measurement data as input and provide the wear category, the wear assessment, a wear prediction, and / or the remaining tool life as output.Additionally, the input, especially when a remaining tool life is to be output, can include information about the intended further use of the tool (Which workpiece is to be machined? In which tool application system is the tool to be used and / or with which machining parameters of the tool application system? etc.). The tasks "evaluating the current wear / the current degree of wear" and "predicting future wear / determining the remaining tool life" can be performed by separately trained and executed machine learning algorithms. However, it is also conceivable that at least the tasks "evaluating the current wear / the current degree of wear" and "predicting future wear / determining the remaining tool life" could be combined and performed by a single, appropriately trained machine learning algorithm.

[0017] The specially trained and further trainable machine learning algorithm can be a, in particular based on deep learning or neural networks, ZO 17913 WO

[0018] The machine learning algorithm in question could be a computer vision algorithm, such as a CNN (Convolutional Neural Network) or a Capsule Networks algorithm. The specifically trained and further trainable machine learning algorithm is preferably one specialized for recognizing objects, geometries, or features in image files. Object, geometry, and feature recognition from images is one of the core disciplines of machine learning, so the training and / or application of such algorithms falls within the expertise of a skilled professional (see, among others, https: / / en.wikipedia.org / wiki / Outline_of_object_recognition, last updated: October 30, 2023, 12:14 UTC).For example, one of the well-known CNN machine learning algorithms described in the following publications can be used in the procedure: a) AlexNet: Alex Krizhevsky, Imagenet classification with deep convolutional neural networks, Communications of the ACM 60.6, pg. 84-90 (2017); b) MobileNet: Andrew G. Howard, MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications, CoRR, abs / 1704.04861, (2017); c) Xception: Francois Chollet, Xception: Deep Learning with Depthwise Separable Convolutions, CoRR, abs / 1610.02357, (2016); d) LeCun Y, Bengio Y, Hinton G (2015) Deep learning; Nature 521: 436{444, DOI 10.1038 / naturel4539; e) Lin H, Li B, Wang X, Shu Y, Niu S (2019); Automated defect inspection of LED chip using deep convolutional neural network; J Intell Manuf; 30:2525{2534, DOI 10.1007 / sl0845-018-1415-x; f) Fu G, Sun P, Zhu W, Yang J, Cao Y, Yang MY, Cao Y (2019); A deep-learning-based approach for fast and robust steel surface defects classification; Opt Laser Eng 121 :397{405, DOI 10.1016 / j.optlaseng.2019.05.005; g) Lee KB, Cheon S, Kim CO (2017) A Convolutional Neural Network for Fault Classification and Diagnosis in Semiconductor Manufacturing Processes; IEEE T Semiconduct M 30: 135{142, DOI 10.1109 / TSM.2017.2676245; h) Goncalves DA, Stemmer MR, Pereira M (2020) A convolutional neural network approach on bead geometry estimation for a laser cladding system; Int J Adv Manuf Tech 106: 1811 {1821 , DOI ZO 17913 WO.

[0019] 10.1007 / s00170-019-04669-z; i) Karatas A, Kölsch D, Schmidt S, Eier M, Seewig J (2019) Development of a convolutional autoencoder using deep neural networks for defect detection and generating ideal references for cutting edges; Munich, Germany, DOI 10.1117 / 12.2525882; j) Stahl J, Jauch C (2019) Quick roughness evaluation of cut edges using a convolutional neural network; In: Proceedings SPIE 11172, Munich, Germany, DOI 10.1117 / 12.2519440; or k) a CNN from the open source framework known as “TensorFlow”. Alternative CNN algorithms known to experts, such as Region Proposals (R-CNN, Fast R-CNN, Faster R-CNN), Detectron, Single Shot MultiBox Detector (SSD), or You Only Look Once (YOLO, for example, in version 9, which was licensable at the time of application), etc., are of course also conceivable. Simple open-source solutions are available for applying many of these machine learning algorithms (see, for example, [reference]).The article from the online encyclopedia “Wikipedia” referenced above. For Capsulate Networks machine learning algorithms, there are, for example, the open-source frameworks “TensorFlow CapsNet”, “PyTorch CapsNet” or Keras “KapsNet”.

[0020] It is conceivable that the training interface is designed to allow further training of the specifically trained and trainable machine learning algorithm using unsupervised learning, supervised learning, reinforcement learning, or a human-in-the-loop approach. The aforementioned training approaches for further training machine learning algorithms—supervised learning, reinforcement learning, and the human-in-the-loop approach—are known to those skilled in the art, so no further explanation regarding their feasibility seems necessary here. The relevant expertise is summarized in particular in the English-language articles of the online encyclopedia "Wikipedia" entitled "Reinforcement learning," "Reinforcement learning from human feedback," and "Supervised learning" (see https: / / en.wikipedia.org / wiki / Reinforcement_learning, accessed October 9, 2017).2024, 03:22 UTC, ZO 17913 WO (https: / / en.wikipedia.org / wiki / Reinforcement_learning_from_human_feedback, accessed: October 9, 2024, 07:11 UTC and https: / / en.wikipedia.org / wiki / Supervised_learning, accessed: August 11, 2024, 13:16 UTC). "Intended" is understood to mean, in particular, specifically programmed, designed, and / or equipped. The fact that an object is intended for a specific function is understood to mean, in particular, that the object fulfills and / or executes this specific function in at least one application and / or operating state. The training interface is preferably also designed to receive training data in the form of pre-built tool wear models. These tool wear models can, for example, originate directly from the manufacturers of the respective tools.The tool manufacturers could also be connected to the tool wear evaluation and / or tool wear prediction network, so that, in particular, the tool wear models could be sent directly to the evaluation and / or prediction computer system or requested or retrieved by the evaluation and / or prediction computer system. It is conceivable that the specially trained and further trainable machine learning algorithm is continuously trained using data received via the training interface, especially the training data and / or synthetic training data, or that data is initially collected via the training interface and then a further training process is initiated periodically or after reaching a minimum data volume.The training interface can be a physical interface module or a software module of the evaluation and / or prediction computer system. The evaluation and / or prediction interface can be a physical interface module or a software program module of the evaluation and / or prediction computer system, stored on a data memory of the evaluation and / or prediction computer system and executable by a processor of the evaluation and / or prediction computer system.Furthermore, it is proposed that either all optical tool measuring devices or a first subset of the optical tool measuring devices in the tool wear evaluation and / or tool wear prediction network each have, preferably at least substantially identical, optical measuring devices that can generate at least substantially the same measurement conditions, such as measurement perspective, measurement distance, camera saturation setting, image resolution, illumination intensity, illumination directions, etc. This advantageously allows for a particularly high quality of wear analysis. It also advantageously enables particularly effective further training of the specially trained and further trainable machine learning algorithm.The optical tool measuring devices, or the first part thereof, could, for example, be some or all of the measuring devices of the "pomBasic," "smile," "venturion," or "genius" types distributed by Zoller at the time of submission of this text. Furthermore, it is conceivable that some of the optical tool measuring devices, or the first part thereof, are other standalone measuring stations or devices mounted on or integrated into tool application systems, which are at least capable of providing essentially identical measuring conditions to the aforementioned devices.Alternatively or additionally, some of the optical tool measuring instruments or the first part of the optical tool measuring instruments may be other measuring instruments whose measuring devices have been adapted, modified or supplemented so that at least essentially the same measuring conditions can be achieved with them as with the aforementioned devices.

[0021] Furthermore, if a second part of the optical tool measuring devices, in particular all optical tool measuring devices not belonging to the first part of the optical tool measuring devices, of the tool wear evaluation and / or tool wear prediction network are intended to generate synthetic training data for the specially trained and further trainable machine learning algorithm, in particular for further training of the specially trained and further trainable machine learning algorithm, from individual optical measurement data whose measurement conditions are in particular ZO 17913 WO different from the aforementioned at least essentially the same measurement conditions, by means of transformations and / or simulations,Since the resulting measurements (after the transformation and / or simulation) correspond to the same measurement conditions as those generated by the optical tool measuring devices of the first part of the tool wear evaluation and / or tool wear prediction network, a particularly high degree of flexibility can be advantageously achieved and / or a particularly large tool wear evaluation and / or tool wear prediction network can be created. Advantageously, a particularly large number of different optical tool measuring devices, especially independent of their manufacturers, can be integrated into the tool wear evaluation and / or tool wear prediction network. The transformations and / or simulations are preferably software-based. The transformations and / or simulations include, in particular, at least perspective changes.Changes in brightness, contrast, and / or size in the optical measurement data, particularly in the tools depicted therein. Specifically, any mention of the term "training data" in this document can be understood as "real measured training data and / or synthetic training data."

[0022] Furthermore, it is proposed that at least one of the optical tool measuring devices be integrated with one of the tool application systems. This allows for particularly high user-friendliness. Advantageously, this enables direct, fast, and / or simple interaction with the evaluation and / or prediction interface of the evaluation and / or prediction computer system and / or with the training interface of the evaluation and / or prediction computer system. ZO 17913 WO

[0023] In particular, the optical tool measuring device integrated with the tool application system can directly generate an image of the tool clamped in the tool application system and send the resulting optical measurement data to the evaluation and / or prediction interface or to the training interface. In particular, another of the optical tool measuring devices of the tool wear evaluation and / or tool wear prediction network is a standalone device, which is designed and arranged separately from any tool application system.

[0024] If one of the wear indicators, specifically intended for training the specially trained and further trainable machine learning algorithm, specifies whether the corresponding, at least partially worn tool is already part of a previously transmitted training dataset and what has happened to the tool in the meantime, then wear prediction and / or service life prediction can be continuously optimized, and thus made increasingly accurate. In particular, this allows the specially trained and further trainable machine learning algorithm to learn, through further training, which wear phenomena occur after which processes on the corresponding tools, e.g., when and with what intensity.Advantageously, the specially trained and further trainable machine learning algorithm can learn to what extent the manufacturer's specifications for tools are accurate, in particular which tools exhibit slower or faster wear than predicted and / or promised by the manufacturer. The relevant wear information can include, for example, the number of machining hours completed, the type and / or duration of machining tasks performed, the type and / or number of workpieces and / or materials machined, and / or the machining parameters used by the tool application equipment. ZO 17913 WO.

[0025] Furthermore, it is proposed that the evaluation and / or prediction computer system include an incentive module designed to identify the successful completion of a further training process using training data received via the training interface, as well as the source of this training data, and to execute a reward transaction. This allows for the advantageous improvement of the specifically trained and further trainable machine learning algorithm. A large amount of training data can be generated, which can lead to continuous improvement of the analysis and / or prediction results.Advantageously, participants in the tool wear evaluation and / or tool wear prediction network can be efficiently and effectively motivated to provide their existing data to the evaluation and / or prediction computer system or to measure existing tools in order to generate further training data. The incentive module is preferably a software program module of the evaluation and / or prediction computer system, stored on the data memory of the system and executable by its processor. In particular, the incentive module is designed to monitor the training interface, capture incoming training data and the associated data sources, and / or monitor further training of the specifically trained and trainable machine learning algorithm to record successful training processes.In particular, the incentive module is designed to send a reward item to the source of the training data via a reward transaction. The reward transaction can be a purely electronic data transfer or it can also include the transfer of physical objects. The reward item could be purely electronic or at least partially physical. The reward transaction is preferably directed to an address associated with / assigned to the source, e.g., a physical address, an internet address, or a digital address within the tool wear evaluation and / or tool wear ZO 17913 WO system.

[0026] Prediction network and / or a crypto address of a distributed ledger technology (DLT), such as a blockchain.

[0027] Furthermore, it is proposed that the reward transaction grants the source credit for a number of future wear evaluations and / or wear predictions (especially free of charge), a number of reputation points (particularly within a quality rating system of the tool wear evaluation and / or tool wear prediction network, preferably publicly accessible), and / or a badge. This can advantageously create a particularly effective incentive to submit and / or generate training data. It can also advantageously improve the specifically trained and further trainable machine learning algorithm. The reputation points can be represented by a star rating, a point rating, a number of points, a monetary (fiat or crypto) reward, a quality score, or the like. It is also conceivable that collected reputation points could be converted into real-world rewards, such as...Maintenance of optical tool measuring devices and / or tool application systems, upgrades of optical tool measuring devices and / or tool application systems, digital fiat money or crypto money, spare and / or wear part deliveries, discounts on future purchases of optical tool measuring devices and / or tool application systems, functional or cosmetic add-on components for optical tool measuring devices and / or tool application systems, physical or digital certificates, or the like, etc. The credit for future wear evaluations and / or wear predictions could be implemented in the form of a digital account to which the credit would be added and from which credit would be deducted for each wear evaluation and / or wear prediction performed. The badge could be a physical badge or a digital badge. Certain attributes could be assigned to the badge, such as...“High quality of wear information”, “frequent transmission of training data”, “regular transmission of training data”, “high tool type variance in ZO 17913 WO the transmitted training data”, “training data for new type of tool”, “training data for new type of wear”, etc.

[0028] Furthermore, it is proposed that the incentive module be designed to forward at least a portion of the rewards sent by a tool application system and / or an optical tool measuring device for received wear evaluations and / or wear predictions, corresponding to relative reputation point values, to the relevant sources that contributed to the training of the specially trained and further trainable machine learning algorithm. This can advantageously create a particularly effective incentive for submitting and / or generating large amounts of training data. The reward can be monetary or non-monetary. The reward could be structured analogously to reputation points. It is conceivable that the forwarding of a portion of the rewards to a specific source could be limited to a period after the last provision of training data.This could advantageously create a particularly effective incentive for the regular submission of training data. It is conceivable that the forwarding of payments could be organized and / or executed via one or more smart contracts built on a distributed ledger technology (DLT). Smart contracts are, in particular, self-executing contracts whose conditions are written in code and preferably stored on a DLT. Smart contracts preferably operate on an "if-then" principle, so that a predefined action is automatically triggered as soon as certain conditions are met. Smart contracts advantageously enable transparent, secure, and automated execution of transactions without the need for an intermediary.For example, the payment of consideration for a wear evaluation and / or a wear prediction could be the "if" condition of the smart contract, and the forwarding of parts of the consideration to sources, e.g., a list to which training data was obtained and which the smart contract has access, could be the "then" condition of the smart contract. ZO 17913 WO.

[0029] Furthermore, it is proposed that the evaluation and / or prediction computer system include a quality module that determines the quality of received training data before it is applied to the specifically trained and further trainable machine learning algorithm. This module then controls the reward transaction of the incentive module based on the determined quality level of the corresponding training data. This allows for the advantageous achievement of high-quality training data, which in turn leads to high-quality wear evaluations and / or wear predictions output by the specifically trained and further trainable machine learning algorithm. For example, a source could be awarded more reputation points for a set of training data the higher its quality. High-quality training data is characterized in particular by good, meaningful, and / or sharp image data.High-quality training data is characterized in particular by complete and / or comprehensive wear information. High-quality training data is also characterized in particular by verifiable accuracy of this wear information. The quality module is specifically a software program module stored on the data memory of the evaluation and / or prediction computer system and executable by the processor of the evaluation and / or prediction computer system. The quality module is designed to read and analyze the received training data and output an analysis result that includes a quality assessment.

[0030] Additionally, it is proposed that the evaluation and / or prediction computer system includes a plausibility module which checks the plausibility of received wear information from an at least partially worn tool, preferably before its application as training data for the specifically trained and further trainable machine learning algorithm, for example by comparison with older training data from the same tool or by comparison with values ​​from other tools of the same source or other sources to identify ZO 17913 WO

[0031] Outliers and / or input errors. This can advantageously lead to high-quality training data, which in turn also leads to high-quality wear analyses and / or wear predictions output by the specially trained and further trainable machine learning algorithm. It is conceivable that the plausibility module controls the reward transaction of the incentive module depending on the determined plausibility of the corresponding training data. It is also conceivable that the plausibility module prevents training of the specially trained and further trainable machine learning algorithm with certain training data if the plausibility module detects insufficient or questionable plausibility in that specific training data.The plausibility module is, in particular, a software program module stored on the data memory of the evaluation and / or prediction computer system and executable by the processor of the evaluation and / or prediction computer system. The plausibility module is designed to read in and analyze the received training data and output an analysis result that includes a plausibility assessment.

[0032] Furthermore, a method, in particular a computer-implemented method, for training a machine learning algorithm specifically trained and further trainable for wear analysis of tools, in particular tool cutting edges, by means of a tool wear evaluation and / or tool wear prediction network according to one of the preceding claims, comprising at least the following method steps: a) measuring an at least partially worn tool by means of an optical tool measuring device connected to the tool wear evaluation and / or tool wear prediction network for generating optical measurement data, b) locally determining a wear degree of the at least partially worn tool from the optical measurement data, manually, automatically or with the support of a local AI, c) optionally recording a machining history of the tool.d) Creating a training dataset for the specially trained and further trainable machine learning algorithm (ZO 17913 WO), comprising at least the optical measurement data of the at least partially worn tool, the associated wear level, and optionally the associated processing history; e) Transmitting the dataset with the training data to an evaluation and / or prediction computer system comprising the specially trained and further trainable machine learning algorithm via a training interface of the evaluation and / or prediction computer system; f) Optionally checking the plausibility of the wear information in the received training data; g) Optionally checking the quality of the received training data; h) Training, in particular further training, the specially trained and further trainable machine learning algorithm using the training data; and i) Optionally executing a reward transaction.This approach is proposed. This allows for the advantageous achievement of high-quality training data, which in turn advantageously leads to particularly high-quality wear evaluations and / or wear predictions output by the specially trained and further trainable machine learning algorithm. If the local determination of the wear degree is performed by a local AI, this local AI, which can be, for example, a computer vision machine learning algorithm based on deep learning or neural networks, such as a CNN (Convolutional Neural Network) machine learning algorithm or a Capsule Networks machine learning algorithm, preferably runs locally on a computing unit assigned to the optical tool measuring device that generates the corresponding optical measurement data.or on a control system of the optical tool measuring device that generates the corresponding optical measurement data. The degree of wear can, for example, be a wear category such as "no wear," "low wear," "beginning wear," "medium wear," "high wear," and / or "completely worn." The machining history can include, among other things, operating times, workpieces machined, machining parameters used, and / or programs. ZO 17913 WO,

[0033] Additionally, a method, in particular a computer-implemented method, for evaluating and / or predicting tool wear by means of a tool wear evaluation and / or tool wear prediction network is described, comprising: a plurality of optical tool measuring devices intended for measuring tools at least once, which are located at at least two geographically distant locations; a plurality of tool application systems, such as machine tools, CNC machining centers, robotic systems, tool management systems, etc., which apply measured tools, wherein at least one of the optical tool measuring devices is geographically co-located with one or more of the tool application systems; and wherein preferably at least one of the optical tool measuring devices is located at a location far removed from each of the tool application systems.and a centrally localized or delocalized evaluation and / or prediction computer system, wherein at least the optical tool measuring devices and the tool application systems are networked with the evaluation and / or prediction computer system directly or indirectly, e.g. via the optical tool measuring device colocated with one of the tool application systems, wherein the evaluation and / or prediction computer system comprises a machine learning algorithm specifically trained and further trainable for wear analysis of tools, in particular cutting edges of the tools,wherein training data in the form of optical measurement data of at least partially worn tools in combination with one or more wear information are received from the optical tool measuring devices via a training interface of the evaluation and / or prediction computer system for further training of the machine learning algorithm, wherein optical measurement data of a tool used or to be used in one of the tool application systems are received from one of the optical tool measuring devices geographically colocated with this tool application system via a prediction interface of the evaluation and / or prediction computer system, whereupon a wear analysis is performed by the specially trained and further trainable machine learning algorithm for an evaluation of the wear of the tool and / or for a prediction of the wear of the tool, in particular a remaining service life of the tool,The wear analysis is performed, and the results are sent back to the relevant tool application system and / or optical tool measuring device. This allows for high-quality and informative tool wear assessments. Wear information from many different locations and tool applications can be collected, enabling optimized and / or particularly precise wear predictions and / or analyses. A particularly powerful AI can be generated and used for future analyses. The prediction and / or analysis results improve continuously and can adapt quickly and flexibly to changes (e.g., in tool geometries or workpiece machining techniques).

[0034] Furthermore, a centrally localized or delocalized evaluation and / or prediction computer system is proposed, comprising at least one computing unit or computing infrastructure including at least a processor and at least a data storage unit. An operating program is stored on the data storage unit, containing instructions executable by the processor, and in particular a specially trained and further trainable machine learning algorithm for executing one or both of the aforementioned procedures. This advantageously enables high-quality and informative tool wear assessments.

[0035] The tool wear evaluation and / or tool wear prediction network according to the invention, the methods according to the invention and / or the evaluation and / or prediction network according to the invention ZO 17913 WO

[0036] The computer system is not intended to be limited to the application and embodiment described above. In particular, the tool wear evaluation and / or tool wear prediction network, the methods according to the invention, and / or the evaluation and / or prediction computer system according to the invention may, to fulfill a functionality described herein, have a different number of individual elements, components, process steps, and units than the number specified herein.

[0037] Drawings

[0038] Further advantages will become apparent from the following description of the drawings. The drawings illustrate an embodiment of the invention. The drawings, the description, and the claims contain numerous features in combination. A person skilled in the art will expediently consider the features individually and combine them into meaningful further combinations.

[0039] They show:

[0040] Fig. 1 shows a schematic representation of a tool wear evaluation and / or tool wear prediction network.

[0041] Fig. 2 shows a schematic flowchart of a method for training a machine learning algorithm specifically trained and further trainable for tool wear analysis using the tool wear evaluation and / or tool wear prediction network and

[0042] Fig. 3 shows a schematic flowchart of a method for evaluating and / or predicting tool wear using the tool wear evaluation and / or tool wear prediction network.

[0043] Description of the exemplary embodiment

[0044] Figure 1 shows a schematic representation of a tool wear evaluation and / or tool wear prediction network 10. The ZO 17913 WO

[0045] The tool wear evaluation and / or tool wear prediction network 10 comprises a plurality of optical tool measuring devices 12, 14, 16, 18. The optical tool measuring devices 12, 14, 16, 18 are each designed for measuring tools 26, in particular at least the cutting edges of tools 26. The optical tool measuring devices 12, 14, 16, 18 are each located at geographically (far) separated locations. The optical tool measuring devices 12, 14, 16, 18 each have an optical measuring device 34. The optical measuring devices 34 of a first part of the optical tool measuring devices 12, 14, 16 of the tool wear evaluation and / or tool wear prediction network 10 can each generate identical measurement conditions.The optical measuring devices 34 of a first part of the optical tool measuring devices 12, 14, 16 of the tool wear evaluation and / or tool wear prediction network 10 are each at least substantially identical in design. The optical measuring devices 34 of a second part of the optical tool measuring devices 18 of the tool wear evaluation and / or tool wear network 10 are identical in design.

[0046] The prediction network 10 is not capable of generating the same measurement conditions. Instead, it generates different individual optical measurement data of the tools 26. The optical tool measuring devices 18 of the second part are designed to generate synthetic training data from the individual optical measurement data through transformations and / or simulations for a specially trained and further trainable machine learning algorithm of the tool wear evaluation and / or tool wear prediction network 10. The generated synthetic training data corresponds to the same measurement conditions as the measurement conditions of the tool measurements generated / generable with the optical tool measuring devices 12, 14, 16 of the first part of the optical tool measuring devices 12, 14, 16 of the tool wear evaluation and / or tool wear prediction network 10. ZO 17913 WO

[0047] The tool wear evaluation and / or tool wear prediction network 10 comprises a plurality of tool application systems 20, 22, 24. The tool application systems 20, 22, 24 are intended for the application of the measured tools 26 / the tools 26 subjected to a wear analysis. Two of the tool application systems 20, 22 are designed as machine tools by way of example. Another of the tool application systems 24 is designed as a tool management system by way of example. Two of the optical tool measuring devices 12, 14 are geographically co-located with one of the tool application systems 20, 22 by way of example. One of these two optical tool measuring devices 12 is positioned in the immediate vicinity of the associated tool application systems 20, but is designed separately from the associated tool application systems 20.The other of these two optical tool measuring devices 14 is designed as an example integral part of the associated tool application system 22.

[0048] The tool wear evaluation and / or tool wear prediction network 10 comprises an evaluation and / or prediction computer system 28. The evaluation and / or prediction computer system 28, shown by way of example in Figure 1, is centrally located. Alternatively, the evaluation and / or prediction computer system 28 could also be configured in a delocalized manner. The optical tool measuring devices 12, 14, 16, 18 are networked with the evaluation and / or prediction computer system 28 via data transmission. The tool application systems 20, 22, 24 are networked with the evaluation and / or prediction computer system 28 via data transmission. For this purpose, the tool application systems 20, 22 are directly connected to the evaluation and / or prediction computer system 28 via data transmission technology and / or indirectly via an optical tool measuring device 12, 14 co-located with the tool application system 20, 22.The evaluation and / or prediction computer system 28 comprises at least the specially trained and further trainable machine learning algorithm. The specially trained and further trainable machine learning algorithm ZO 17913 WO is specifically trained for wear analysis of tools 26, in particular tool cutting edges. The specially trained and further trainable machine learning algorithm can be continuously further trained using training data.

[0049] The evaluation and / or prediction computer system 28 has a training interface 30. The training interface 30 is designed to receive training data from the optical tool measuring devices 12, 14, 16, 18. The training data comprises optical measurement data of at least partially worn tools 26 in combination with one or more wear information about the tool. The wear information includes details of whether the associated at least partially worn tool 26 is already part of a previously transmitted dataset of training data and what has happened to the tool 26 since the transmission of this older dataset.The wear information can also include the number of machining hours completed, the type and / or duration of machining tasks performed, the type and / or number of workpieces and / or materials machined, and / or machining parameters used by the respective tool application system 20, 22, 24. The training data serves to further train the specially trained and further trainable machine learning algorithm. The evaluation and / or prediction computer system 28 has an evaluation and / or.

[0050] Prediction interface 32. The evaluation and / or prediction interface 32 is designed to receive optical measurement data of a tool 26 used or to be used in one of the tool application systems 20 from one of the optical tool measuring devices 12, which is geographically co-located with this tool application system 20. The evaluation and / or prediction interface 32 is designed to perform a wear analysis using the specially trained and further trainable machine learning algorithm. The wear analysis includes an evaluation of the wear of the tool 26 and / or a prediction of the wear of the tool 26. The wear analysis can also include a ZO 17913 WO

[0051] The prediction of the remaining tool life of the tool 26 is included. The evaluation and / or prediction interface 32 is designed to send the results of the wear analysis back to the tool application system 20 and / or to the associated optical tool measuring device 12.

[0052] The evaluation and / or prediction computer system 28 includes an incentive module 36. The incentive module 36 is designed to identify the successful execution of a further training process based on training data received via the training interface 30 from one of the optical tool measuring devices 12, 14, 16, 18 (=sources). The incentive module 36 is designed to identify the source of this training data. The incentive module 36 is designed to execute a reward transaction, in particular to the source or to a target address associated with the source. The reward transaction assigns a number of reputation points to the source. The reputation points can be part of a publicly accessible or non-publicly accessible quality assessment system of the tool wear evaluation and / or tool wear prediction network 10. However, the reputation points can also have other content or value.

[0053] Alternatively or additionally, the reward transaction assigns the source a credit for a number of future wear evaluations and / or wear predictions. Alternatively or additionally, the reward transaction assigns the source a badge. The incentive module 36 is designed to forward at least a portion of the considerations sent by a tool application system 20, 22, 24 and / or an optical tool measuring device 12, 14, 16, 18 for received wear evaluations and / or wear predictions, e.g., to the evaluation and / or prediction computer system 28, according to the relative number of reputation points of the sources, to the associated sources that contributed to the training of the specially trained and further trainable machine learning algorithm. For this purpose, the evaluation and / or prediction computer system 28 uses a smart contract. ZO 17913 WO

[0054] The evaluation and / or prediction computer system 28 includes a quality module 38. The quality module 38 is designed to determine the quality of the received training data for further training of the specifically trained and trainable machine learning algorithm before it is applied to the specifically trained and trainable machine learning algorithm. The quality module 38 is designed to control the reward transaction(s) of the incentive module 36 depending on a determined quality level of the corresponding training data, in particular to adjust their scope / value. The evaluation and / or prediction computer system 28 includes a plausibility module 40. The plausibility module 40 is designed to check the plausibility of wear information received with training data from a tool 26 that is at least partially worn.The verification is performed before the training data containing the wear information is used to further train the specifically trained and trainable machine learning algorithm. The plausibility check can be performed by comparing the data with older training data from the same tool or by comparing it with values ​​from other tools from the same source or from other sources. The plausibility check serves to identify outliers within the wear information that indicate errors and / or potential input errors within the wear information.

[0055] The evaluation and / or prediction computer system 28 comprises a computing unit 46. The computing unit 46 comprises a processor 42. The computing unit 46 comprises a data memory 44. The specially trained and further trainable machine learning algorithm is stored on the data memory 44. The specially trained and further trainable machine learning algorithm can be executed by means of the processor 42. An operating program is stored on the data memory 44, which contains instructions executable by the processor 42 for the execution of one or more of the procedures 100, 200 described below. The evaluation and / or prediction computer system 28 is intended for the execution of at least part of, and in particular all of, the procedures 100, 200 described below.

[0056] Figure 2 shows a schematic flowchart of a method 100 for training a machine learning algorithm specifically trained and further trainable for the wear analysis of tools 26, in particular their cutting edges, using the tool wear evaluation and / or tool wear prediction network 10. The method 100 is a computer-implemented method. In at least one process step 110, a tool 26 that is at least partially worn is measured using one of the optical tool measuring devices 12, 14, 16, 18 connected to the tool wear evaluation and / or tool wear prediction network 10. Optical measurement data for this tool 26 are generated. In at least one further process step 120, a wear degree of the measured tool 26 is determined from the optical measurement data.The degree of wear is determined locally, in particular by a control unit assigned to the optical tool measuring device 12, 14, 16, 18, which generates the optical measurement data, or by a user of the optical tool measuring device 12, 14, 16, 18, which generates the optical measurement data. The degree of wear can be determined completely manually or automatically in process step 120. Preferably, the degree of wear is determined in process step 120 using a local machine learning algorithm assigned to the optical tool measuring device 12, 14, 16, 18, which generates the optical measurement data, and which is specifically trained for the detection and evaluation of wear characteristics. In at least one optional further process step 130, a machining history of the tool 26 is determined.In at least one further process step 140, a data set with training data is created for the specially trained and further trainable machine learning algorithm of the evaluation and / or prediction computer system 28. The data set includes at least the optical measurement data of the at least partially worn tool 26 and the degree of wear associated with the same tool 26 ZO 17913 WO and optionally the machining history associated with the same tool 26.

[0057] In a further process step 150, the data set containing the training data is transmitted to the evaluation and / or prediction computer system 28, which comprises the specially trained and further trainable machine learning algorithm. The evaluation and / or prediction computer system 28 receives the data set containing the training data via the training interface 30. In a further optional process step 160, the plausibility of the wear information of the received training data is checked by the plausibility module 40 of the evaluation and / or prediction computer system 28. In a further optional process step 170, the quality of the received training data is checked by the quality module 38 of the evaluation and / or prediction computer system 28.In a further process step 180, the specially trained and further trainable machine learning algorithm is trained, and in particular further trained, using the received training data, especially the measured training data and / or the synthetic training data. In an optional further process step 190, one or more of the reward transactions are executed. The described process 100 can be executed multiple times sequentially for all provided datasets with training data. This ensures that the specially trained and further trainable machine learning algorithm is continuously trained, so that its capacity for wear assessment and wear prediction constantly increases and the results become increasingly precise.

[0058] Figure 3 shows a schematic flowchart of a method 200 for evaluating and / or predicting the wear of a tool 26 using the previously described tool wear evaluation and / or tool wear prediction network 10, which comprises the specially trained and further trainable machine learning algorithm that can be continuously trained according to the previously described method 100. ZO 17913 WO

[0059] Method 200 is a computer-implemented method. In at least one method step 210, optical tool measuring devices 12, which are geographically colocated with one of the tool application systems 20, determine optical measurement data of a tool 26 that is currently used or will be used in the future in this tool application system 20. In a further method step 220, this optical measurement data is transmitted to the evaluation and / or prediction computer system 28 for analysis. The transmitted optical measurement data is received by the evaluation and / or prediction interface 32 of the evaluation and / or prediction computer system 28.In an optional further process step 230, the optical tool measuring device 12 or the tool application system 20 transmitting the optical measurement data for analysis makes a payment, in particular in the form of a transaction directed to the evaluation and / or prediction computer system 28. In a further process step 240, the received optical measurement data are evaluated by the specially trained and further trainable machine learning algorithm. The specially trained and further trainable machine learning algorithm performs a wear analysis to evaluate the wear of the tool 26. Alternatively or additionally, the specially trained and further trainable machine learning algorithm performs a wear analysis to predict the wear of the tool 26, in particular the remaining tool life of the tool 26.In a further process step 250, a result of the wear analysis(s) is sent back to the tool application system 20 and / or to the optical tool measuring device 12, from which the optical measurement data to be analyzed originally originated. In a further optional process step 260, part of the consideration is forwarded via a smart contract of a DLT connected to the evaluation and / or prediction computer system 28 to one or more sources that have participated in the further training of the specially trained and further trainable machine learning algorithm by means of the previously described process 100.

[0060] Reference sign

[0061] 10 Tool wear evaluation and / or tool wear prediction network

[0062] 12 Optical tool measuring device

[0063] 14 Optical tool measuring device

[0064] 16 Optical tool measuring device

[0065] 18 Optical tool measuring device

[0066] 20 tool application system

[0067] 22 Tool application system

[0068] 24 Tool application system

[0069] 26 tools

[0070] 28 Evaluation and / or prediction computer system

[0071] 30 Trainer interface

[0072] 32 Evaluation and / or prediction interface

[0073] 34 Optical measuring device

[0074] 36 Incentive Module

[0075] 38 Quality Module

[0076] 40 Plausibility Module

[0077] 42 processor

[0078] 44 Data storage

[0079] 46 computing units

[0080] 100 procedures

[0081] 110 Procedure step

[0082] 120th process step

[0083] 130 Procedure step

[0084] 140 Procedure step

[0085] 150th process step

[0086] 160 Process step

[0087] 170 Procedure step

[0088] 180 process step

[0089] 190 Procedure step ZO 17913 WO

[0090] Proceedings

[0091] Procedure step

[0092] Procedure step

[0093] Procedure step

[0094] Procedure step

[0095] Procedure step

[0096] Procedure step

Claims

ZO 17913 WO October 21, 2025 Claims 1. Tool wear evaluation and / or tool wear prediction network (10), comprising: - a plurality of optical tool measuring devices (12, 14, 16, 18) intended at least for measuring tools (26), which are located at at least two geographically distant locations, - a plurality of tool application systems (20, 22, 24), such as machine tools, CNC machining centers, robot systems, tool management systems, etc., which apply measured tools (26), wherein at least one of the optical tool measuring devices (12, 14, 16, 18) is geographically co-located with one or more of the tool application systems (20, 22, 24), - a centrally localized or delocalized evaluation and / or prediction computer system (28), wherein at least the optical tool measuring devices (12, 14, 16, 18) and the tool application systems (20, 22, 24) are networked with the evaluation and / or prediction computer system (28) directly or indirectly, e.g. via the optical tool measuring device (12, 14) colocated with one of the tool application systems (20, 22, 24), wherein the evaluation and / or prediction computer system (28) comprises at least one machine learning algorithm specifically trained and further trainable for wear analysis of tools (26), in particular tool cutting edges, wherein the evaluation and / or prediction computer system (28) has a training interface (30) which is provided for ZO 17913 WO Training data in the form of optical measurement data of at least partially worn tools (26) in combination with one or more wear information for further training of the specially trained and further trainable machine learning algorithm from the optical tool measuring devices (12, 14, 16, 18), and wherein the evaluation and / or prediction computer system (28) has an evaluation and / or prediction interface (32) which is intended to receive optical measurement data of a tool (26) used or to be used in one of the tool application systems (20) from one of the optical tool measuring devices (12) which is geographically co-located with this tool application system (20),to receive and, following at least one wear analysis performed by the specially trained and further trainable machine learning algorithm, to send an evaluation of the wear of the tool (26) and / or a prediction of the wear of the tool (26), in particular a remaining service life of the tool (26), back to the tool application system (20) and / or to the optical tool measuring device (12).

2. Tool wear evaluation and / or tool wear prediction network (10) according to claim 1, characterized in that either all optical tool measuring devices (12, 14, 16, 18) or a first part of the optical tool measuring devices (12, 14, 16) of the tool wear evaluation and / or tool wear prediction network (10) each have, preferably at least substantially identical, optical measuring devices (34) which can generate at least substantially the same measurement conditions, such as measurement perspective, measurement distance, camera saturation setting, image resolution, illumination intensity, illumination directions, etc. ZO 17913 WO 3. Tool wear evaluation and / or tool wear prediction network (10) according to claim 2, characterized in that a second part of the optical tool measuring devices (18), in particular all optical tool measuring devices (18) not belonging to the first part of the optical tool measuring devices (12, 14, 16), of the tool wear evaluation and / or tool wear prediction network (10) are provided to generate synthetic training data for the specially trained and further trainable machine learning algorithm from individual optical measurement data by means of transformations / simulations, which correspond to the same measurement conditions as the tool measurements generated with the optical tool measuring devices (12, 14, 16) of the first part of the optical tool measuring devices (12, 14, 16) of the tool wear evaluation and / or tool wear prediction network (10).

4. Tool wear evaluation and / or tool wear prediction network (10) according to one of the preceding claims, characterized in that at least one of the optical tool measuring devices (14) is integrally formed with one of the tool application systems (22).

5. Tool wear evaluation and / or tool wear prediction network (10) according to one of the preceding claims, characterized in that one of the wear information indicates whether the associated at least partially worn tool (26) is already part of a previously transmitted data set of training data and what has happened to the tool (26) in the meantime. ZO 17913 WO 6. Tool wear evaluation and / or tool wear prediction network (10) according to one of the preceding claims, characterized in that the evaluation and / or prediction computer system (28) comprises an incentive module (36) which is designed to identify a successful execution of a further training process based on training data received via the trainer interface (30) and a source of this training data and to execute a reward transaction.

7. Tool wear evaluation and / or tool wear prediction network (10) according to claim 6, characterized in that the reward transaction of the source - a loan for a number of future wear assessments and / or wear predictions, - a number of reputation points, in particular of a quality assessment system of the tool wear evaluation and / or tool wear prediction network (10), preferably publicly accessible, and / or - assigns a badge.

8. Tool wear evaluation and / or tool wear prediction network (10) according to claim 7, characterized in that the incentive module (36) is provided to forward considerations sent by a tool application system (20, 22, 24) and / or an optical tool measuring device (12, 14, 16, 18) for obtained wear evaluations and / or wear predictions, at least in part, according to relative reputation point numbers, to the associated sources that contributed to the training of the specially trained and further trainable machine learning algorithm. ZO 17913 WO 9. Tool wear evaluation and / or tool wear prediction network (10) according to one of claims 6 to 8, characterized in that the evaluation and / or prediction computer system (28) comprises a quality module (38) which determines the quality of received training data before application to the specially trained and further trainable machine learning algorithm and controls the reward transaction of the incentive module (36) depending on a determined quality level of the corresponding training data.

10. Tool wear evaluation and / or tool wear prediction network (10) according to one of the preceding claims, characterized in that the evaluation and / or prediction computer system (28) comprises a plausibility module (40) which checks the plausibility of received wear information of an at least partially worn tool (26), preferably before its application as training data for the specially trained and further trainable machine learning algorithm, for example by comparison with older training data from the same tool (26) or by comparison with values ​​of other tools (26) from the same source or other sources to identify outliers and / or input errors.

11. Method (100), in particular a computer-implemented method, for training a machine learning algorithm specifically trained and further trainable for wear analysis of tools (26), in particular of tool cutting edges, by means of a tool wear evaluation and / or tool wear prediction network (10) according to one of the preceding claims, comprising the method steps: - Measuring a tool that is at least partially worn (26) using a tool wear evaluation and / or ZO 17913 WO Tool wear prediction network (10) connected optical tool measuring device (12, 14, 16, 18) for generating optical measurement data, - local manual, automated or locally AI-assisted determination of the degree of wear of the at least partially worn tool (26) from the optical measurement data, - optional recording of a tool's editing history (26), - Creating a training data set for the specially trained and further trainable machine learning algorithm, comprising at least the optical measurement data of the at least partially worn tool (26), the associated degree of wear and optionally the associated machining history, - Sending the data set containing the training data to an evaluation and / or prediction computer system (28) comprising the specially trained and further trainable machine learning algorithm, which receives the data set via a training interface (30) of the evaluation and / or prediction computer system (28), - Optionally, check the plausibility of the wear information in the received training data. - optionally check the quality of the received training data, - Training, and in particular further training, of the specially trained and further trainable machine learning algorithm using the training data and - Optionally, execute a reward transaction.

12. Method (200), in particular a computer-implemented method, for evaluating and / or predicting the wear of a tool (26) using a tool wear evaluation and / or tool wear prediction network (10), in particular according to one of claims 1 to 10, comprising: - a plurality of at least one tool for measuring (26) ZO 17913 WO provided optical tool measuring devices (12, 14, 16, 18), which are located at at least two geographically distant locations, - a plurality of tool application systems (20, 22, 24), such as machine tools, CNC machining centers, robot systems, tool management systems, etc., which apply measured tools (26), wherein at least one of the optical tool measuring devices (12, 14, 16, 18) is geographically co-located with one or more of the tool application systems (20, 22, 24), and - a centrally localized or delocalized evaluation and / or prediction computer system (28), wherein at least the optical tool measuring devices (12, 14, 16, 18) and the tool application systems (20, 22, 24) are networked with the evaluation and / or prediction computer system (28) directly or indirectly, e.g., via the optical tool measuring device (12, 14) colocated with one of the tool application systems (20, 22, 24), wherein the evaluation and / or prediction computer system (28) comprises a machine learning algorithm specifically trained and further trainable for wear analysis of tools (26), in particular tool cutting edges, wherein the optical tool measuring devices (12, 14, 16, 24) receive data via a training interface (30) of the evaluation and / or prediction computer system (28).18) Training data in the form of optical measurement data of at least partially worn tools (26) in combination with one or more wear information are received for further training of the specially trained and further trainable machine learning algorithm, wherein, via an evaluation and / or prediction interface (32) of the evaluation and / or prediction computer system (28), one of the tools application systems (20) is geographically co-located, optical tool measuring devices (12) receive optical measurement data of a tool (26) used or to be used in this tool application system (20), whereupon a wear analysis is performed by the specially trained and further trainable machine learning algorithm to evaluate the wear of the tool. (26) and / or to predict the wear of the tool (26), in particular the remaining service life of the tool (26), and wherein a result of the wear analysis is sent back to the corresponding tool application system (20) and / or to the corresponding optical tool measuring device (12).

13. Centrally localized or delocalized evaluation and / or prediction computer system (28) comprising at least one computing unit (46) or computing infrastructure comprising at least one processor (42) and at least one data storage (44), wherein an operating program is stored on the data storage (44) which comprises instructions executable by the processor (42), and in particular a specially trained and further trainable machine learning algorithm, for executing a method (100, 200) according to one of claims 11 or 12.

Images