An induction coil unit, a method for monitoring the induction heating process of an induction coil unit, and a self-learning system for monitoring the induction heating process and a method for training the self-learning system.

Abstract

The present invention provides an induction coil unit, a method for monitoring the induction heating process of the induction coil unit, and a self-learning system for monitoring the induction heating process, as well as a method for training the self-learning system. [Solution] Provide at least one, in particular several, different induction process characteristics, in particular coil current and / or coil voltage and / or input current and / or input voltage and / or intermediate circuit current and / or intermediate circuit voltage. Provide at least one or more different characteristics to a learning system, in particular a neural network, which, using at least one or more different characteristics, determines the temperature characteristics of the sleeve portion inserted into the induction coil and / or the sleeve characteristics of the sleeve portion inserted into the induction coil and / or the time characteristics of the heating process of the sleeve portion inserted into the induction coil.

Description

Technical Field

[0001] The present invention relates to an induction coil unit and a method for monitoring an induction heating process of the induction coil unit. Further, the present invention relates to a self-learning system for monitoring an induction heating process and a method for training the self-learning system.

Background Art

[0002] Induction coil units of that kind are known from the prior art of Patent Document 1, for example.

[0003] These known induction coil units use an alternating magnetic field that can be generated by an induction coil and eddy currents induced in a tool holder inserted into the induction coil of the induction coil unit to thermally expand the tool holder so that a tool can be inserted into the expanded tool holder. The tool is firmly and symmetrically held in the tool holder after the cooling process of the tool holder. This process is briefly also referred to as induction shrink fitting of the tool into the tool holder and is known as such.

[0004] For a more detailed explanation regarding the technical and operational background, refer to the above patent application.

[0005] However, in such known induction coil units, for their efficient operation, i.e., induction shrink fitting of the tool into the tool holder, especially when heating the tool holder, the induction coil unit needs to be individually set for each tool holder held at that time with respect to various operating parameters or shrink fitting parameters, such as heating time (or shrink fitting / heating frequency, shrink fitting / heating temperature, maximum time of the heating process, or (shrink fitting / heating) energy). This requires a high degree of manual operation, and as a result, the cycle time for replacing tools in different types of tool holders may be significantly lengthened. Further, manual operation is always a potential cause of errors.

[0006] On the other hand, if such settings are not made for the currently held tool holder, or if the settings are incorrect, the expected eddy currents may not be properly induced in the tool holder, potentially leading to inefficient operation of the induction coil unit. In particular, under adverse conditions such as excessively long heating times, overheating could lead to damage and even destruction of the tool holder.

[0007] Furthermore, self-learning systems are also known.

[0008] A self-learning system is a system that learns from experience or data and can thereby automatically improve without requiring explicit programming instructions. This system adapts its functional form based on new information to optimize its performance.

[0009] A common example of a self-learning system is a machine learning algorithm (such as a neural network, decision tree, or support vector machine) that can recognize patterns from large amounts of data and make predictions or decisions. Unlike traditional programs where all rules and procedures are pre-defined, self-learning systems improve through continuous training, and the more data they are faced with, the more accurate or efficient their predictions become.

[0010] Self-learning systems can perform tasks such as image classification, text translation, or even decision-making in dynamic environments, without continuous human intervention.

[0011] The example neural network shown above, as is evident from conventional technologies, is a machine learning model inspired by the functional structure of the human brain.

[0012] Such neural networks consist of interconnected "neurons" organized into multiple layers to process data and recognize patterns. This type of network is particularly well-suited for tasks such as image recognition, language processing, translation, or prediction.

[0013] The main components of a neural network are neurons, which are composed of multiple layers (input layer, output layer, and hidden layer), as described above. A neuron (node) is the basic processing unit of a network. Each neuron receives input data, performs calculations, and passes the result to the neuron in the next layer.

[0014] The input layer receives input data (e.g., images or text) and passes it to the next layer. The output layer outputs the final result or prediction of the model, such as image classification. Hidden layers are located between the input and output layers and perform the main calculations. In these layers, complex patterns and features are extracted from the input data.

[0015] Each connection between neurons has a weight that determines the signal transmission strength from one neuron to the next. Furthermore, there is also bias, which shifts the neuronal outcomes and increases the flexibility of the model.

[0016] After a neuron processes input data, an activation function is often applied to determine whether the neuron is activated (i.e., "fires"). Examples of activation functions include the sigmoid function, ReLU (Rectified Linear Unit), and Tanh.

[0017] Neural networks improve through training. Large amounts of data are used to fit weights and biases, enabling the network to make the most accurate predictions or classifications possible. This process is typically carried out using a method called backpropagation. In this method, the error between the network's predictions and the actual results is propagated backward within the network to fit the weights.

[0018] Neural networks, especially deep networks (so-called deep learning networks), can solve extremely complex tasks with high accuracy, but this requires a large amount of data and enormous computing resources.

[0019] Decision trees are also known as self-learning systems. In particular, in artificial intelligence, especially machine learning, decision trees are a preferred algorithm for classification and regression.

[0020] A decision tree is a systematic model for representing and analyzing decision-making processes. It represents decisions and their possible outcomes in a tree-like structure. The starting point is the start node, from which various branches branch out, each representing a different action option or event.

[0021] In this process, decision nodes (usually represented as rectangles) where conscious choices are made are distinguished from random nodes (usually represented as circles) where various events can occur with a given probability. The endpoint of the tree often represents an outcome that can be evaluated in terms of cost, benefit, or profit.

[0022] Decision trees are used to make complex decisions transparent, to compare options, and to select the best course of action based on probabilities and expected outcomes.

[0023] Decision trees are trained, just like other self-learning systems such as neural networks (see above). For this purpose, available and well-known training methods / algorithms include, for example, top-down induction (recursive partitioning), C4.5 and its successors (e.g., CART, ID3), bagging (bootstrap aggregation), random forests, or boosting (e.g., AdaBoost, gradient boosting).

[0024] Furthermore, implementation methods for designing and optimizing decision tree structures, such as Scikit-learn, are also known.

[0025] Scikit-learn provides a particularly good implementation of the CART algorithm (classification and regression trees) (mentioned above), and this algorithm can be used for both classification and regression problems.

[0026] With Scikit-learn, decision trees can be easily trained by passing data and target variables to corresponding classes (e.g., decision tree classification or decision tree regression). In doing so, Scikit-learn calculates criteria such as Gini impurity or information gain to find the best split.

Prior Art Documents

Patent Documents

[0027]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Patent Document 5

Patent Document 6

Summary of the Invention

Problems to be Solved by the Invention

[0028] The problem of the present invention is to provide an induction coil unit, a method for monitoring an induction heating process of the induction coil unit, a self-learning system for monitoring the induction heating process, and training of such a self-learning system. This can overcome the drawbacks of this type of induction coil unit known from the prior art, and in doing so, the induction coil unit is particularly highly automated, thereby enabling highly reliable operation with high process reliability at shorter cycle times with high operating reliability.

Means for Solving the Problems

[0029] This problem is solved by an induction coil unit, a method for monitoring the induction heating process of an induction coil unit, a self-learning system for monitoring the induction heating process, and a method for training such a self-learning system. These have the features of their respective independent claims. An advantageous development of the present invention is the subject of the dependent claims and the following specification, relating to an induction coil unit, a method for controlling the induction heating process of an induction coil unit, a self-learning system for monitoring the induction heating process, and a method for training such a self-learning system.

[0030] Terms such as up, down, front, back, left, or right, used as needed, should be understood according to their common understanding and with reference to the drawings of this application, unless otherwise defined. Terms such as radial and axial should be understood in relation to the central axis and axis of symmetry of the parts / components described herein, and with reference to the accompanying drawings, unless otherwise defined.

[0031] The term "basically," when used, can be interpreted (according to the Supreme Court's understanding) as meaning "substantial, to a considerable extent." The possible deviations from accuracy implied by this term may occur unintentionally (i.e., without functional reason), such as due to manufacturing or assembly tolerances.

[0032] An induction coil unit and a method for monitoring the induction heating process of the induction coil unit are provided, wherein the induction coil unit includes an induction coil into which the sleeve portion of a tool holder can be inserted.

[0033] A key core of the present invention is a self-learning system, particularly a neural network or decision tree, wherein at least one, in particular, multiple different characteristic quantities of the induction process are provided or can be provided, particularly current and / or voltage, such as coil current and / or coil voltage and / or input current and / or input voltage and / or intermediate circuit current and / or intermediate circuit voltage.

[0034] Furthermore, a self-learning system, particularly a neural network or decision tree, is provided with at least one characteristic quantity, such as the instantaneous change in coil current, and / or the instantaneous change in intermediate circuit current, and / or the instantaneous rate of change of coil current, and / or the instantaneous rate of change of intermediate circuit current, and / or the absolute coil voltage, and / or the absolute coil current, and / or the energy from the coil current that has flowed since the start of heating, and / or the energy from the intermediate circuit current that has flowed since the start of heating, and / or the integral of the instantaneous coil current that has flowed since the start of heating (effective current), and / or the integral of the intermediate circuit current that has flowed since the start of heating, and / or the heating time since the start of heating, and / or the instantaneous inductance of the entire system consisting of the induction coil or the induction coil and sleeve portion, and / or the instantaneous phase angle between the instantaneous coil current (effective current) and the coil voltage driving it.

[0035] Further such characteristics (provided / possible to provide to self-learning systems, particularly neural networks or decision trees) may be the time (duration) over which the coil current or intermediate circuit current undergoes a predetermined change, for example, by increasing. For example, this is the time (duration) required for the coil current or intermediate circuit current to increase from a first (lower) limit (e.g., 0A) to a second (upper) limit (e.g., a configurable maximum current value). This occurs, for example, when a test pulse (described later for test pulses) is applied to the coil current or intermediate circuit current.

[0036] This self-learning system determines the monitored characteristic using at least one or more different characteristic quantities (and, if necessary, at least one additional characteristic quantity). For example, the temperature characteristic quantity of the sleeve portion inserted into the induction coil, and / or the sleeve characteristic quantity of the sleeve portion inserted into the induction coil, and / or the time characteristic quantity of the heating process of the sleeve portion inserted into the induction coil.

[0037] As used in this invention, the term "characteristic quantity" refers to a (process) quantity or variable, either a (one-dimensional) value or a (multidimensional) vector, particularly a variable that changes over time, such as current, voltage, temperature, inductance, or energy.

[0038] Self-learning systems, or neural networks / decision trees, are optimized when a large number of hidden layers are used, especially 10 or more hidden layers, especially 15 or more hidden layers, especially 20 or more hidden layers, especially 25 or more hidden layers, especially 30 or more hidden layers, and especially 32 hidden layers.

[0039] Furthermore, to further improve self-learning systems or neural networks, it is advantageous to use the ReLU function and / or the sigmoid function as activation functions.

[0040] A self-learning system or neural network / decision tree can be trained using the above characteristics to determine the above monitoring characteristics, such as the temperature characteristic of the sleeve portion inserted into the induction coil, and / or the sleeve characteristic of the sleeve portion inserted into the induction coil, and / or the time characteristic of the heating process of the sleeve portion inserted into the induction coil (method of training a self-learning system).

[0041] Typically, training is carried out using so-called training data (or in the form of so-called training data vectors), using as much of such data as possible, for example, under the use of characteristics provided to a self-learning system or neural network / decision tree, in particular, characteristics provided to a self-learning system or neural network / decision tree (see: at least one or more different characteristics, or at least one further characteristic, in particular current and / or voltage, in particular coil current and / or coil voltage and / or input current and / or input voltage and / or intermediate circuit current and / or intermediate circuit voltage and / or in particular instantaneous change in coil current and / or instantaneous rate of change of coil current and / or absolute coil voltage and / or absolute coil current and / or energy from coil current flowing since heating started The process is performed using the integral of the instantaneous coil current (active current) that has flowed since the start of heating, and / or the heating time since the start of heating, and / or the instantaneous inductance of the induction coil, or the entire system consisting of the induction coil and the sleeve portion, and / or the instantaneous phase angle between the instantaneous coil current (active current) and the coil voltage driving it, and / or the time (duration) over which the coil current or intermediate circuit current undergoes a predetermined change, such as an increasing time, and in particular, also in connection with monitoring characteristics provided by a self-learning system or neural network / decision tree (see: temperature characteristics of the sleeve portion inserted into the induction coil, and / or sleeve characteristics of the sleeve portion inserted into the induction coil, and / or time characteristics of the heating process of the sleeve portion inserted into the induction coil) ("training data / vectors").

[0042] This training data is determined by a defined heating process, and in particular from the values ​​measured therein. Specifically, it can be obtained by a heating process performed on a measuring cone, and during or between this "defined" heating process, the thus set (measured) characteristic quantities can be obtained along with their corresponding (measured) monitoring characteristic quantities.

[0043] For example, these (measured) training data can be acquired with a "resolution" of 5, 10, or 20 values ​​per second during heating.

[0044] Furthermore, relevant monitoring characteristics can be obtained through user feedback and / or user interaction. In other words, here, a self-learning system, or neural network / decision tree, can be trained using user feedback and / or user interaction.

[0045] To improve the performance of self-learning systems or neural networks, training using backpropagation methods is provided.

[0046] To improve self-learning systems or neural networks, they can also be retrained, particularly with the use of user feedback and / or user interaction.

[0047] Nevertheless, in training, it is important to incorporate multiple data sources to enable comprehensive modeling. This may include actual measurement data, simulation scenarios, or synthetic data to prepare the system for rare or extreme situations.

[0048] This data can be used to prepare self-learning systems, or neural networks / decision trees, for a variety of scenarios that rarely occur in reality. Such data helps improve the robustness and generalization ability of the models.

[0049] This invention is based on the recognition that complex processes are difficult or impossible to describe through (accurate) analysis, and therefore the process quantities that can be determined by analysis are not sufficiently precise. Consequently, control based on such analysis carries operational safety risks. In other words, in the case of a tool holder using (induction) heating, purely "analytically constructed" monitoring and control of the (induction) heating process of the sleeve portion or the tool holder based on analytically determined quantities (e.g., Patent Document 2 by the applicant) may pose safety risks.

[0050] Self-learning systems, such as neural networks or decision trees, provide a suitable means for more accurately representing or describing complex processes, such as the (induction) heating process of a sleeve, by learning the process (based on training data). The self-learning system or neural network / decision tree creates a kind of "image" of the process / heating process (heating step), and this image can be used to determine process parameters, such as monitoring characteristics, for monitoring and control. Based on this, the process can be monitored, controlled, and / or adjusted.

[0051] In short, the self-learning system, or neural network / decision tree, of the present invention generates highly reliable process quantities (referred to as monitoring characteristic quantities in this specification) with high accuracy, and based on these, it is possible to perform (process) monitoring and control for operational stability and error avoidance (see the limit value comparison for blocking described below).

[0052] Furthermore, the self-learning system, or neural network / decision tree approach of the present invention, offers the potential for predictive use. It can monitor the current process state, identify abnormalities and deviations from normal conditions, and even predict future developments. This is particularly useful for proactive "maintenance" and optimization of operating parameters. Potential failures can be detected early and corrected (by adjustments as necessary) before they lead to costly downtime. This improves reliability and availability.

[0053] The combination of (conventional) "analysis-based" and "neural" monitoring (and control) can make the monitoring of induction heating processes more reliable.

[0054] Furthermore, the present invention offers the advantage of increasing the level of automation. When the present invention is incorporated, for example, into automatic induction heating of a tool holder in an automated tool changer, the characteristics in the self-learning system or neural network can be automatically acquired during the process, and since the self-learning system or neural network / decision tree determines the monitored characteristics, manual operation by the user is not required during the process. In other words, the process can be reliably executed with a shorter cycle time and high operational reliability.

[0055] method In a method for controlling the induction heating process of an induction coil unit, for example, a shrink-fitting device, for example, a sleeve portion of a tool holder inserted into the induction coil of the induction coil unit, is configured to provide at least one characteristic quantity, in particular several different characteristic quantities, of the induction heating process.

[0056] To further improve and shape the heating process by a self-learning system, it is preferable that at least one further characteristic quantity is provided, such as the instantaneous change in coil current, and / or the instantaneous change in intermediate circuit current, and / or the instantaneous rate of change of coil current, and / or the instantaneous rate of change of intermediate circuit current, and / or the absolute coil voltage, and / or the absolute coil current, and / or the energy from the coil current that has flowed since the start of heating, and / or the energy from the intermediate circuit current that has flowed since the start of heating, and / or the integral of the instantaneous coil current (active current) that has flowed since the start of heating, and / or the integral of the intermediate circuit current that has flowed since the start of heating, and / or the heating time from the start of heating, and / or the instantaneous inductance of the induction coil, or the entire system consisting of the induction coil and the sleeve portion, and / or the instantaneous phase angle between the instantaneous coil current (active current) and the coil voltage driving it.

[0057] Furthermore, the characteristic quantity provided / can be provided may be the time (duration) over which the coil current or intermediate circuit current undergoes a predetermined change, for example, by increasing. For example, this may be the time (duration) required for the coil current or intermediate circuit current to increase from a first (lower) limit (e.g., 0A) to a second (upper) limit (e.g., the maximum configurable current). This is, for example, the case of the coil current or intermediate circuit current in an applied test impulse (a test impulse will be discussed later).

[0058] Nevertheless, such characteristic quantities may be quantities determined using at least one or more different characteristic quantities, such as integrals, derivatives, or similar methods, for example, quantities like energy as the integral of the coil current (active current).

[0059] It is preferable to pass the characteristics provided to a self-learning system through a digital smoothing filter, such as a Kalman filter, before feeding them into the learning system. This filter is particularly suitable for real-time data processing because it continuously updates predictions and incorporates measurements to generate more accurate estimates. The filter is ideal for dynamic systems such as induction processes where the state may change rapidly. The use of an adaptive algorithm can optimize the performance of the filter by dynamically adapting the adaptive algorithm to the changing conditions of the heating process.

[0060] "Provide" may mean that at least one or more different characteristics and / or at least one further characteristics are measured before and / or during the heating process, particularly during the heating process, or that values ​​are provided that are specifically stored or calculated for at least one or more different characteristics and / or at least one further characteristics.

[0061] Furthermore, characteristic quantities can also be "provided" by measurement using corresponding sensors or sensor technologies, particularly sensors with IoT technology.

[0062] At least one or more different characteristics, or these and at least one additional characteristic, are supplied to the learning system.

[0063] The learning system determines, using the supplied characteristic quantities, monitoring characteristic quantities such as the temperature characteristic quantity of the sleeve portion inserted into the induction coil, and / or the sleeve characteristic quantity of the sleeve portion inserted into the induction coil, and / or the time characteristic quantity of the heating process of the sleeve portion inserted into the induction coil.

[0064] For example, the temperature characteristic quantity could be the temperature of the sleeve portion of the tool holder. The sleeve characteristic quantity could be the shape information of the sleeve portion or the tool holder, in particular the size of the tool holder, or information classifying (or describing) the sleeve portion or the tool holder. The time characteristic quantity could be the residual heating time.

[0065] The heating process is monitored using monitoring characteristics such as temperature characteristics and / or sleeve characteristics and / or time characteristics.

[0066] Furthermore, if the monitored characteristic variable, and / or the characteristic variable supplied to the self-learning system, reaches a configurable limit, or exceeds or falls below that limit, it is possible to terminate (or at least modify) the heating of the sleeve portion. This ensures that the heating process is carried out reliably.

[0067] It is considered preferable to set or readjust heating parameters and / or cutoff parameters based on established monitoring characteristics, and to monitor, execute, or control the heating process accordingly. It is also possible to terminate (or at least modify) the heating of the sleeve portion when such parameters reach, exceed, or fall below their settable limits.

[0068] Furthermore, to further enhance the level of heating automation, it may be possible to provide a system that, before starting, automatically determines the shape of the sleeve portion inserted into the induction coil, such as its outer diameter, thereby "recognizing" the inserted sleeve portion.

[0069] This is described, for example, in Patent Documents 3, 4, and 5. The contents of those documents are the subject of this application.

[0070] If the inserted sleeve portion is recognized, heating parameters can be set for the inserted sleeve portion or its heating process, such as provided characteristic quantities or characteristic quantities supplied to the self-learning system.

[0071] Such heating parameters may include, in particular, the frequency of shrink-fitting / heating, and / or the shrink-fitting / heating temperature, as well as the duration of the heating process, and / or the maximum duration of the heating process, and / or the energy (current integral). From these, limits for the heating process (especially for "neural" monitoring and / or "analytical" monitoring) can also be defined.

[0072] In a further preferable development, it may be offered that multiple decision trees be used, rather than just one.

[0073] In particular, it may be preferable to use multiple decision trees having the same structure. Each decision tree can be trained with the same output quantity or output parameters (and the final output quantity can include all output quantities (e.g., according to the share)). This allows all combinations of decision trees to improve the reliability or predictive accuracy of the monitored characteristic.

[0074] unit According to the present invention, the induction coil unit for implementing this method is equipped with a self-learning system according to the present invention.

[0075] The self-learning system determines the monitoring characteristics based on the use of the provided characteristics, as described above.

[0076] Furthermore, in the induction coil configuration unit, at least one current detector and / or voltage detector can be provided in the intermediate circuit of the frequency converter that supplies electrical energy to the coil circuit, and / or in the coil circuit having the induction coil.

[0077] The induction coil unit is particularly preferred if it includes a comparator for comparing at least one of the above-mentioned characteristic quantities, such as a monitoring characteristic quantity and / or a characteristic quantity supplied to a self-learning system, with a configurable limit value.

[0078] In comparison, it is preferable for the induction coil unit to have an automatic shut-off device configured to shut off (or at least change) the heating process by the induction coil when the above-mentioned configurable limit value is reached, or when the limit value is exceeded or below.

[0079] Furthermore, in order to further improve the induction coil unit, it may be provided that the induction coil unit be equipped with at least two self-learning systems, particularly operating in parallel, that determine multiple different monitoring characteristics, such as temperature characteristics and time characteristics, or sleeve characteristics and time characteristics, or sleeve characteristics and temperature characteristics.

[0080] To further improve the operational reliability of the induction coil unit, it may be preferable for the circuit to include at least one power semiconductor element, in particular at least one bipolar transistor (IGBT) having an insulated gate electrode, and / or a metal-oxide-semiconductor field-effect transistor (MOSFET) (which have good conduction characteristics, high reverse voltage, and robustness, and can be controlled with virtually no power).

[0081] This method and induction coil unit enable nearly fully automated operation, namely the induction shrink-fitting of tools into tool holders, particularly the heating of the tool holders, for each tool holder inserted into the induction coil unit. This eliminates the need for manual operation parameter setting, saving time that was previously required, and also ensures high standards of operational reliability and tolerance through automation, thereby guaranteeing the unit's specified operation. Furthermore, this method and induction coil unit enable efficient prevention of overheating of the heated / expanded tool holders. Power output can be optimized to reduce energy consumption and increase efficiency.

[0082] The advantageous embodiments of the present invention described herein include many features that are grouped together in individual dependent claims. However, these features can also be considered individually for practical purposes and combined into meaningful further combinations.

[0083] In particular, these features can be used individually or in any suitable combination with the method and / or apparatus according to the present invention.

[0084] Furthermore, even if certain terms are used in the singular form or in combination with numbers in the specification or claims, the scope of the invention relating to these terms is not limited to the singular form or the respective numbers. Moreover, the words "ein" or "eine" should be understood as indefinite articles, not as numbers.

[0085] The characteristics, features, advantages, and methods for achieving the above invention will be made clearer and easier to understand by being described in more detail in the following description of embodiments of the invention and in relation to the drawings / illustrations (in the drawings / illustrations, identical components / parts and functions are indicated by the same reference numerals).

[0086] The examples are for illustrative purposes only and do not limit the invention to any combination of features described therein, nor to any functional features. Furthermore, appropriate features of each example may be explicitly separated and examined, deleted from one example, added to another, or combined with any claim. [Brief explanation of the drawing]

[0087] [Figure 1] This is a central longitudinal cross-sectional view showing an induction coil unit according to one embodiment. [Figure 2] This is a circuit diagram of a circuit that supplies power to an induction coil that can be used in the induction coil unit shown in Figure 1. [Figure 3]This figure shows the neural network used to monitor the heating process of the induction coil unit shown in Figure 1. [Figure 4] This figure shows the change in coil current for different tool holders in the induction coil unit shown in Figure 1 in response to a test pulse. [Figure 5] This figure shows a decision tree used to monitor the heating process in the induction coil unit shown in Figure 1. [Modes for carrying out the invention]

[0088] Automatic heating control of the shrink-fit process using a shrink-fitting device / Monitoring of the induction heating process using a self-learning system (neural network or decision tree) (Figures 1-2 and 4-3, 5) Figure 1 shows the basic structure of an induction coil unit, which is called a shrink-fit device (or shrink-fit equipment) due to its function (explained below).

[0089] Induction coil unit / shrink-fit device As shown in Figure 1, the shrink-fitting device comprises an induction coil 1 having individual windings 2, into which a tool holder 4 is inserted to shrink-fit or unshrink-fit the holding shaft H of a tool W (in this case, a milling cutter) within the sleeve portion HP.

[0090] The operating principles underlying shrink-fitting and shrink-release are described in detail in Patent Document 6. The contents of that document are the subject of this application.

[0091] A first jacket 3 made of a material that is electrically nonconductive and magnetically conductive is provided on the outer circumference of the induction coil 1.

[0092] Typically, the first jacket 3 consists of ferrite, metal powder, or metal sintered material, and its individual particles are electrically insulated from each other. In this way, the overall structure is basically magnetically conductive and electrically nonconductive.

[0093] The first jacket 3 is also completely closed in the circumferential direction, that is, it completely covers the circumferential surface of the induction coil 1, and therefore, theoretically, there is no “magnetic gap” at all, except for insignificant local penetrations such as individual holes and / or small local holes.

[0094] As further shown in Figure 1, in the shrink-fit device, shielding with magnetically conductive and electrically non-conductive materials is not limited to the first jacket 3.

[0095] Instead, magnetic covers 3a and 3b made of the above material are attached to at least one, more preferably both, end faces of the first jacket 3, and are usually in contact with the first jacket 3.

[0096] On the end face of the induction coil 1 away from the tool holder 4, a magnetic cover 3a is formed, preferably as a pole piece that is replaceable entirely or more preferably partially, i.e., as an annular structure having a central opening that forms a through hole 7 for the tool W to be attached or removed.

[0097] On the end face of the induction coil 1 facing the tool holder 4, a magnetic cover 3b is preferably formed as a flat annular disc that completely covers the windings of the induction coil 1 and has a central through hole for the sleeve portion HP.

[0098] To further improve shielding, as shown in Figure 1, the induction coil 1 and its first jacket 3 are surrounded on their outer circumference by a second jacket 9, and the first jacket 3 and the second jacket 9 are in contact with each other, ideally, they are in contact over most or all of their facing circumferential surfaces.

[0099] This second jacket 9 is made of a magnetically nonconductive and electrically conductive material, such as aluminum.

[0100] In this embodiment, "electrical conductivity" refers not merely to a material that is locally, or so to speak, "particulate," conductive, but to a material that allows the formation of eddy currents within a relevant range.

[0101] The second characteristic of the jacket 9 is that it is preferably designed to be radially thick so that, under the influence of the stray magnetic field of the induction coil 1, eddy currents are generated inside it that weaken undesirable stray magnetic fields.

[0102] Furthermore, the second jacket 9 is surrounded by a power semiconductor element 10, which will be described later. The power semiconductor element 10 is directly positioned within the void 11 (approximately) on the outer periphery of the second jacket 9.

[0103] These power semiconductor elements 10 have two large main surfaces and four small side surfaces. Preferably, the large main surfaces are four times larger than each of the side surfaces.

[0104] The power semiconductor element 10 is positioned such that one of its large main surfaces is in thermal conductive contact with the second jacket 9, typically at the outer periphery of the second jacket 9. In this configuration, the large main surface of the power semiconductor element 10 is bonded to the circumferential surface of the second jacket 9 using a thermally conductive adhesive.

[0105] Each power semiconductor element 10 has various power supply terminals.

[0106] Furthermore, as shown in Figure 1, capacitors 14a and 14b are grouped around the outer periphery of the induction coil 1.

[0107] Capacitor 14a is preferably a smoothing capacitor that is a direct component of the power circuit, and capacitor 14b is preferably a capacitor for an oscillating circuit that is also a direct component of the power circuit.

[0108] In this embodiment, in order to electrically connect capacitors 14a and 14b, a plurality of circuit boards 15a and 15b are provided, each surrounding the outer circumference of the induction coil 1.

[0109] These circuit boards 15a and 15b each preferably form an annular disk. Each circuit board 15a and 15b is preferably made of FR4 or a similar material commonly used for circuit boards.

[0110] As can be seen from Figure 1, the rotational symmetry axes of the two circuit boards 15a and 15b, which are implemented as annular substrate disks, are coaxial with the longitudinal axis L of the induction coil (and tool holder 4 / tool ​​W) in this embodiment.

[0111] The upper of the two circuit boards 15a has a smoothing capacitor 14a mounted on it, and its connection flag either penetrates the upper circuit board 15a or is connected to the upper circuit board 15a using SMD technology, so that the smoothing capacitor 14a hangs downward from the upper circuit board 15a.

[0112] The two lower circuit boards 15b are configured accordingly, and the vibration circuit capacitor 14b protrudes upward from its circuit board.

[0113] In short, the power semiconductor element 10 forms a first virtual cylinder surrounding the induction coil 1, the capacitors 14a and 14b form a second virtual cylinder surrounding the first virtual cylinder, the capacitors 14a and 14b, which are less sensitive to stray magnetic fields, form a virtual outer cylinder, while the power semiconductor element 10, which requires an installation space where the influence of stray magnetic fields is minimized, forms a virtual inner cylinder.

[0114] Furthermore, as shown in Figure 1, the induction coil 1 is not "completely wound" along its longitudinal axis L over its entire length. Instead, in this embodiment, the induction coil 1 is composed of two winding packages, which are, in principle, cylindrical. These each form the end faces of the induction coil 1. These end faces are spaced at least 1.5 times, for example, greater than the spread of each winding package in the longitudinal axis L direction of the induction coil 1.

[0115] Such an induction coil 1 contributes to reducing reactive power because it lacks a "central region" winding, which, while not strictly necessary from the standpoint of heating the sleeve portion HP of the tool holder as effectively as possible, tends to generate additional reactive power without making a significant contribution to heating if present.

[0116] To minimize the losses in induction coil 1, a circuit detailed in Figure 2 is provided.

[0117] circuit As shown in Figure 2, this circuit has an oscillating circuit SKS for that purpose (see Figure 2).

[0118] In the SKS vibration circuit, most of the required energy is generated by periodic (at high frequencies) reciprocating vibrations between the induction coil 1 and the capacitor units 14a and 14b. This means that only the energy lost due to heat generation and other power losses in the SKS vibration circuit needs to be resupplied during each period, or periodically. This eliminates the extremely large losses of conventional designs.

[0119] As shown in Figure 2, the power electronics supplying power to induction coil 1 are powered on the input side by a commonly available commercial power supply NST, which is 400V / 50Hz in Europe (three-phase AC, 3f) (corresponding values ​​in other countries).

[0120] As shown in Figure 2, the current supplied from the commercial power source is converted to DC by the rectifier G21 and smoothed by one or more smoothing capacitors 14a (not shown).

[0121] This DC current powers the actual vibration circuit SKS, as shown in Figure 2.

[0122] The backbone of the vibration circuit SKS forms a power semiconductor element 10, a vibration circuit capacitor 14b, and an induction coil 1 for shrink-fitting and shrink-release.

[0123] The vibration circuit SKS is controlled or regulated by the control electronic device SEK, which is supplied with DC power from the rectifier G.

[0124] The power semiconductor element 10 is preferably realized by an "insulated gate bipolar transistor," or IGBT-type transistor.

[0125] The control electronic device SEK switches the power semiconductor element 10 / IGBT according to a frequency that specifies the operating frequency set in the vibration circuit SKS.

[0126] It is important that the SKS vibration circuit never resonates and operates precisely.

[0127] Resonant operation could quickly destroy the power semiconductor element 10 due to voltage spikes. Instead, the control electronic device SEK is configured to drive the power electronic device, or its vibration circuit SKS, within a predetermined operating range close to the system's resonance or natural frequency.

[0128] Preferably, the vibration circuit is controlled or adjusted (by the control device 20) so that 0.9 ≤ cosφ ≤ 0.99 is satisfied. In particular, a value in the range of 0.95 ≤ cosφ ≤ 0.98 is preferred. This avoids voltage spikes, thereby further promoting miniaturization.

[0129] Shrink-fit / heating process monitoring - overheat protection To ensure the shrink-fitting device operates with the required operational safety and automate it as much as possible, the shrink-fitting device is equipped with an automatic heating control / adjustment device that enables automated shrink-fitting operations.

[0130] This heating control / adjustment device is implemented by a corresponding control or adjustment device 20 within the shrink-fitting device, which also includes overheat protection for the tool holder 4 currently inserted into the induction coil for shrink-fitting (to prevent the tool holder currently being shrink-fitted from being damaged by overheating).

[0131] Important for the control device or adjustment device 20 and for overheat protection are the currents and voltages generated in the circuit, namely, in this embodiment, in particular the coil current M-SpA, the coil voltage M-SpV, and the intermediate circuit current M-EA and / or the intermediate circuit voltage M-EV.

[0132] To measure the coil current (M-SpA), coil voltage (M-SpV), intermediate circuit current (M-EA), and intermediate circuit voltage (M-EV), the circuit, as shown in Figure 2, is equipped with a corresponding number of current / voltage measuring instruments M-SpA (coil current), M-SpV (coil voltage), or M-EA (intermediate circuit current), or M-EV (intermediate circuit voltage), which are incorporated into the circuit at the corresponding locations and in the manner shown in the figure. These measuring instruments provide the corresponding measurements at a sampling rate of 10 values / second.

[0133] If the measured quantity, or the progress of the induction coil used in shrink fitting, also depends on the temperature of the sleeve portion into which the tool holder is inserted, then this measured quantity, or these measured quantities, can be advantageously utilized in a neural-based (see Figure 3, neural network 50; Figure 5, decision tree 60) for automated monitoring or heating control to improve the safety of the shrink fitting device by avoiding "manual" sources of error, since the process is automated.

[0134] In other words, the above quantities (measurable by the measuring device), coil current M-SpA, coil voltage M-SpV, and intermediate circuit current M-EA and / or intermediate circuit voltage M-EV (along with other characteristic quantities, as described later) are input to the neural network 50 (see Figure 3) (or decision tree 60, Figure 5) as input quantities, and in the former exemplary embodiment, the neural network 50 outputs the temperature HP-Temp of the sleeve portion of the inserted tool holder as a possible output quantity.

[0135] Furthermore, the neural network 50 (see Figure 3) also outputs a residual heating time HP-Rt as another possible output quantity. This represents the amount of time that the heating process must still be performed from the predetermined / current time to ensure "optimal shrink-fitting," based on the residual heating time HP-Rt output from the neural network 50 at a predetermined / current time t. "Optimal shrink-fitting" is characterized by being non-destructive, allowing the tool to be removed and inserted, and being performed in the shortest possible time.

[0136] In this embodiment, the characteristic quantities supplied to the neural network 50 as further input quantities are so-called delta quantities, which can be derived or calculated from the above current and voltage values, namely the delta coil current Δ-M-SpA, the delta intermediate circuit current Δ-M-EA, the delta energy Δ-Ener, and the delta Kalman Δ-Kalm.

[0137] A further characteristic quantity, the delta coil current Δ-M-SpA, is based on the derivative of the coil current. Another characteristic quantity, the delta intermediate circuit current Δ-M-EA, is based on the derivative of the intermediate circuit current. A further characteristic quantity, the delta energy Δ-Ener, is the amount of energy inserted into the tool holder or sleeve portion being shrink-fitted during heating (see the integral of the coil current). A further characteristic quantity, the delta Kalman Δ-Kalm, is the output of the Kalman filter supplied by the delta intermediate circuit current Δ-M-EA.

[0138] Induction heating is controlled by the control device or adjustment device 20 and is performed until the first of the above output levels reaches its respective cutoff criterion. That is, it is performed until the temperature HP-Temp of the sleeve portion of the currently inserted tool holder reaches a predetermined maximum temperature, or until the residual heating time HP-t decreases to "zero residual time". After that, the induction coil is automatically turned off.

[0139] This "neural" control or adjustment 20, or overheat protection, can be further superimposed by analytical control or adjustment 20, or overheat protection, by using additional characteristic quantities, such as a predetermined maximum heating time and / or the amount of energy (inserted into the tool holder or sleeve portion currently being shrink-fitted). For this purpose, in the heating process, the actual current heating time is measured, and the amount of energy (integral of the instantaneous coil current (active current) since the start of heating) inserted into the sleeve portion up to the present time (continuously) is determined. If the (measured) actual current heating time (for the tool holder currently being shrink-fitted) reaches an individually set limit, or if the amount of energy reaches an individually set limit, the heating process is terminated, or the induction coil is shut off.

[0140] Other characteristic quantities, such as inductance or phase angle, as mentioned in this application, can also be used as further cutoff criteria in appropriate ways.

[0141] Neural Network 50 Figure 3 shows the neural network 50.

[0142] The neural network 50 has an input layer 51, 32 hidden layers 52 in this case, and an output layer 53.

[0143] The input layer 51, which has eight (input) neurons, is supplied with the above-mentioned (characteristic) quantities: coil current M-SpA, coil voltage M-SpV, intermediate circuit current M-EA and intermediate circuit voltage M-EV, as well as other characteristic quantities: delta coil current Δ-M-SpA, delta intermediate circuit current Δ-M-EA, delta energy and delta Kalman.

[0144] The neural network 50 determines the obtainable monitoring characteristics in the output layer 53, which has two (output) neurons: the temperature HP-T of the insertion sleeve portion of the tool holder, and the current residual heating time HP-Rt.

[0145] ReLU and sigmoid are used as activation functions.

[0146] Training a neural network in 50 steps To train the neural network 50, measurements are taken on a defined measurement cone, or on a measurement cone that has been "shrink-fitted" by a shrink-fitting device during the heating process. In addition to the measured current and voltage, a temperature sensor also measures the temperature of the sleeve portion of the inserted measurement cone. The current residual heating time is calculated by subtracting the heating time elapsed since the start of the current "shrink-fitting" from the maximum heating time (the current "shrink-fitting").

[0147] The measured values ​​(current, voltage, temperature) and current residual heating time are collected for the (total) heating time of the performed "shrink-fit" operation, according to the sampling rate (10 values / second in this embodiment).

[0148] The numerous training data vectors formed from this are used to train the neural network.

[0149] Alternative: Instead of the neural network 50 described in this embodiment (which provides two output / monitoring characteristic quantities: the temperature HP-T of the inserted sleeve portion of the tool holder, and the current residual heating time HP-Rt), two parallel-operating neural networks can also be used.

[0150] Both are supplied with the above-mentioned (characteristic) quantities: coil current M-SpA, coil voltage M-SpV, intermediate circuit current M-EA and intermediate circuit voltage M-EV, as well as other characteristic quantities: delta coil current Δ-M-SpA, delta intermediate circuit current Δ-M-EA, delta energy and delta Kalman, respectively. The first neural network outputs the temperature HP-T of the inserted sleeve portion of the tool holder, and the second neural network outputs the current residual heating time HP-Rt.

[0151] The training of the two networks is performed according to the "overall network" (neural network 50).

[0152] (Automatic) recognition of the tool holder / sleeve portion currently inserted into the shrink-fitting device or its induction coil. Before heating begins of the sleeve portion (HP, see Figure 1) inserted into the induction coil (1, see Figures 1 and 2), the shape or outer diameter of the sleeve portion inserted into the induction coil is automatically determined, and the inserted sleeve portion is "recognized" as a result.

[0153] In particular, a test pulse with known current amount, current waveform, frequency, and duration is applied to the sleeve portion inserted into the induction coil before the start of the actual induction heating process. For this test pulse, the time / current curve of the sleeve portion inserted into the induction coil is determined, and this time / current curve, as a whole, is used for the sleeve portion inserted into the induction coil as a magnetic fingerprint (determining its shape).

[0154] This is described, for example, in Patent Documents 3, 4, and 5. Their contents are the subject of this application ("referred by reference").

[0155] Figure 4 illustrates the time / current curves (curves 1, 2, and 3) for three (different) sleeve sections or tool holders (tool holders 1, 2, and 3) inserted into the induction coil for a defined test pulse. As can be seen from Figure 4, the time / current curves have distinct and specific progressions for each of the three (different) sleeve sections or tool holders inserted into the induction coil, and therefore, the individual sleeve sections or tool holders can be inferred from the progression of the time / current curves.

[0156] If the inserted sleeve portion is recognized, the (pre-set) heating parameters, particularly (individual) shrink-fit / heating frequencies, can be determined for the inserted sleeve portion or its heating process.

[0157] Alternative: The neural network is also trained on the shape or outer diameter of the sleeve portion inserted into the induction coil, and therefore can determine the shape or outer diameter of the sleeve portion inserted into the induction coil.

[0158] The heating process of the inserted sleeve portion is initiated by its individual shrink-fit / heating frequency, at which time the coil current, coil voltage, intermediate circuit current, and intermediate circuit voltage are measured simultaneously with the start of power supply to the induction coil (see circuit in Figure 2).

[0159] The aforementioned "neural" monitoring and control enable nearly fully automatic or automated shrink-fitting. This eliminates the need for manual intervention in setting operating parameters, saving time that was previously required. Furthermore, the automatic / automated operation ensures that high standards for operational safety and tolerances are met, guaranteeing that the unit operates as specified. It also provides efficient protection against overheating of the sleeve portion.

[0160] Further use of test pulses in heating control (characteristic quantities Δt and ∫) (Figure 4) As described above, Figure 4 illustrates the time / current curves for three (different) sleeve sections inserted into the induction coil, or the tool holder, for a defined test pulse.

[0161] As shown in Figure 4, from such a time / current curve relating to a given tool holder / sleeve portion, two further characteristic quantities relating to the tool holder / sleeve portion can be determined.

[0162] -Characteristic quantity Δt The characteristic quantity Δt represents the time it takes for the coil current to reach its maximum value (see Figure 4, "max.") when a test pulse is applied (see Figure 4, Δt1, Δt2, Δt3).

[0163] -Characteristic quantity ∫ The characteristic quantity ∫ represents the area under the time / current curve at a defined point in time, or a defined, or even pre-configurable event (in this embodiment, the area under the time / current curve at the time of the test pulse (see Figure 4, Test Pulse 2), during the period when the coil current decreases after the coil current reaches its maximum value (see Figure 4, Test Pulse 2)).

[0164] Alternatively, the characteristic quantity ∫ can be the entire area below the time / current curve during the test pulse.

[0165] As shown in Figure 4, both characteristic quantities Δt and ∫ are specific to each tool holder / sleeve portion and are therefore suitable for representing or defining a specific, given tool holder. This allows for the execution of a predetermined control function in the induction heating process of a shrink-fitting apparatus for the tool holder / sleeve portion represented by the two characteristic quantities Δt and ∫.

[0166] This is based on a database of coil current measurements of test pulses in a given tool holder (representing nearly the entire product range) for configurable coil positions during a possible (successful) shrink-fit process in a shrink-fit device.

[0167] Based on this, for each tool holder, two characteristic quantities Δt and ∫ are determined for every possible (successful) coil shrink-fit position for each tool holder. Each of these defined shrink-fit processes can be associated with its respective (successful) shrink-fit / heating parameter set (a "dataset" consisting of "tool holder, characteristic quantity Δt and characteristic quantity ∫, coil position, shrink-fit / heating parameters").

[0168] This database can be used for predetermined control functions during the induction heating process in shrink-fitting equipment.

[0169] - Control of coil settings using characteristic quantities Δt and ∫ When controlling the coil setting in the shrink-fitting process of a known shrink-fit target tool holder, the tool holder is inserted into the shrink-fitting device, and the induction coil is set to a predetermined (successful) coil shrink-fitting position for this known tool holder.

[0170] A test pulse is applied, the coil current is measured, and two characteristic quantities, Δt and ∫, are determined. The database is then tested to see if there is a dataset with characteristic quantities Δt and ∫ that (approximately) match those of the test pulse.

[0171] In that case, the coil shrink-fit position setting is considered correct. If a corresponding dataset is not found in the database, the coil shrink-fit position setting is reported as incorrect.

[0172] - Determination of shrink-fit parameters using characteristic variable Δt and ∫ In this embodiment, an (unknown) tool holder is inserted into the shrink-fitting device, and the guide coil is set to a coil shrink-fitting position (which is (from the perspective of an experienced operator) is (successful)).

[0173] A test pulse is applied, and the coil current is measured to determine two characteristic quantities, Δt and ∫. In the database, the dataset containing the two characteristic quantities Δt and ∫ that best match those of the test pulse is precisely identified.

[0174] After identifying this "suitable" dataset, we can obtain shrink-fit parameters for (unknown) tool holders from this dataset and perform (successful) shrink-fitting of the tool holders based on these characteristics.

[0175] Nevertheless, the two characteristics Δt and ∫ (either individually or in combination) can also be supplied as input quantities to a given neural network, in addition to the input quantities already described. The training data / dataset can be adjusted accordingly.

[0176] Monitoring of shrink-fitting / heating processes - Overheat protection using decision tree 60 (Figure 5) Figure 5 shows (exemplary) decision tree 60.

[0177] This decision tree 60 has three decision levels 61a, b, and c, as well as an outcome level 62. Structurally, this decision tree provides eight outcome values ​​EGW1, EGW2, ... EGW8 (outcome level 62), which are used to correct the assumed initial / starting value AW of the temperature HP-T of the insertion sleeve portion of the tool holder to the (predicted) temperature HP-T of the insertion sleeve portion of the tool holder 63.

[0178] In this case, 100 decision trees 60, each with the same structure, are used, and their correction values ​​are reflected in the correction by 1% for each of the 100 decision trees 60.

[0179] Thus, the initial / starting temperature HP-T of the sleeve portion in which the tool holder is inserted, corrected using all decision trees or all their correction values, becomes the (predicted) temperature HP-T of the sleeve portion in which the tool holder is inserted (relative to a predetermined heating point) at that time.

[0180] The decision tree 60 is supplied with 10 (measured) characteristic quantities KG of the induction process and 5 other characteristic quantities wKG (derived from them) for each heating time t. For example, the above-mentioned characteristic quantities are the coil current M-SpA, coil voltage M-SpV, and intermediate circuit current M-EA and intermediate circuit voltage M-EV, as well as the other characteristic quantities are the delta coil current Δ-M-SpA, delta intermediate circuit current Δ-M-EA, and delta energy and delta Kalman. Each decision tree 60 makes a decision for its node and provides the corresponding result values ​​EGW1, EGW2, ... EGW8.

[0181] The 30ms sampling rate in the induction process provides characteristic quantities for the decision tree 60 during the induction process, and the decision tree 60 determines the temperature HP-T of the inserted sleeve portion of the tool holder at that point, as described above.

[0182] If the determined measured temperature HP-T of the inserted sleeve portion of the tool holder exceeds a predetermined limit temperature, the induction process is terminated.

[0183] Training of 60 decision trees Decision trees are trained using CART / Scikit-learn, with the structure specified for three decision levels and one outcome level.

[0184] The training data is formed as a 16-row vector containing the 10 characteristic quantities in KG and 5 additional characteristic quantities in wKG, as well as correction values. This training generates numerous decision trees (each with its trained node parameter (i.e., a predetermined characteristic value) and its decision value) and correction values, and 100 "best" decision trees are selected from among them. The initial / starting values ​​of the temperature HP-T of the sleeve portion inserted into the tool holder are also selected.

[0185] For example, one of the selection criteria for the chosen decision trees 60 could be differences in node parameters. The greater the differences among the trained decision trees 60, the more accurate and reliable the prediction of temperature values ​​across the range of possible sleeve components will be.

[0186] Normalized characteristics can be used as needed. This may also become apparent as a result of the training process.

[0187] Although the present invention has been described in detail by preferred embodiments, the present invention is not limited to the disclosed embodiments, and other modifications can be derived without departing from the scope of protection of the present invention. [Explanation of Symbols]

[0188] 1. Induction coil 2 windings 3. First jacket 3a, 3b Magnetic cover 4 Tool holders 9. Second jacket 10 Power semiconductor devices 11 Vacant part 14a, 14b Capacitors 15a, 15b Circuit board 20 Control / Adjustment Devices 21 Rectifier G 50 Neural Networks 51 Input Layer 52 Hidden Layers 53 Output Layer 60 Decision Trees H retaining shaft HP sleeve section HP-T sleeve temperature M-SpA coil current M-SpV Coil Voltage M-EA intermediate circuit current M-EV Intermediate Circuit Voltage SEK Control Electronics SKS vibration circuit

Claims

1. A method for monitoring an induction heating process for an induction coil unit, wherein the induction coil unit has a sleeve portion of a tool holder inserted into the induction coil of the induction coil unit, To provide at least one, in particular several, different, characteristic quantities of an inductive process, in particular, coil current and / or coil voltage and / or input current and / or input voltage and / or intermediate circuit current and / or intermediate circuit voltage, At least one or more different characteristic quantities are supplied to the learning system, particularly to the neural network, and in doing so, the learning system determines, using at least one or more different characteristic quantities, a monitoring characteristic quantity such as the temperature characteristic quantity of the sleeve portion inserted into the induction coil, and / or the sleeve characteristic quantity of the sleeve portion inserted into the induction coil, and / or the time characteristic quantity of the heating process of the sleeve portion inserted into the induction coil. A method for monitoring an induction heating process, characterized by monitoring the heating process using the monitoring characteristic quantities, such as the temperature characteristic quantity and / or the sleeve characteristic quantity and / or the time characteristic quantity.

2. To provide at least one further characteristic quantity such as the instantaneous change in the coil current and / or the intermediate circuit current, and / or the instantaneous rate of change of the coil current and / or the intermediate circuit current, and / or the absolute coil voltage, and / or the absolute coil current, and / or the energy from the coil current and / or intermediate circuit current that has flowed since the start of heating, and / or the integral of the instantaneous coil current (active current) and / or intermediate circuit current that has flowed since the start of heating, and / or the heating time from the start of heating, and / or the instantaneous inductance of the induction coil or the entire system consisting of the induction coil and the sleeve portion, and / or the instantaneous phase angle between the instantaneous coil current (active current) and the coil voltage driving it, and / or the time (duration) over which the coil current or the intermediate circuit current undergoes a predetermined change, for example, an increase, and / or A method for monitoring an induction heating process according to claim 1, characterized by supplying the at least one further characteristic quantity to a learning system that determines the monitoring characteristic quantity using the at least one further characteristic quantity.

3. A method for monitoring an induction heating process according to claim 1 or 2, characterized in that the at least one or more different characteristic quantities and / or the at least one further characteristic quantity are provided by measurement before and / or during the heating process, particularly during the heating process, or by providing values ​​for the at least one or more different characteristic quantities and / or the at least one further characteristic quantity, particularly by providing stored or calculated values.

4. A method for monitoring an induction heating process according to any one of claims 1 to 3, characterized in that the characteristic quantities provided to the self-learning system are passed through a digital smoothing filter, such as a Kalman filter or a normalization filter, before being supplied to the learning system.

5. A method for monitoring an induction heating process according to any one of claims 1 to 4, characterized in that when the monitored characteristic quantity and / or the characteristic quantity supplied to the self-learning system reaches a settable limit value, or exceeds or falls below that limit value, the heating of the sleeve portion is terminated (or at least modified).

6. A method for monitoring an induction heating process according to any one of claims 1 to 5, characterized in that the temperature characteristic quantity is the temperature of the sleeve portion of the tool holder, and / or the sleeve characteristic quantity is morphological information of the sleeve portion or the tool holder, in particular the size of the tool holder, or information classifying the sleeve portion or the tool holder, and / or the time characteristic quantity is the residual heating time.

7. A method for monitoring an induction heating process according to any one of claims 1 to 6, characterized by setting or readjusting heating parameters and / or cutoff parameters using determined monitoring characteristics.

8. An induction coil unit for carrying out the method according to any one of claims 1 to 7, comprising an induction coil into which the sleeve portion of a tool holder can be inserted, An induction coil unit comprising a self-learning system that determines monitoring characteristics such as a temperature characteristic quantity of a sleeve portion inserted into an induction coil, and / or a sleeve characteristic quantity of the sleeve portion inserted into the induction coil, and / or a time characteristic quantity for the heating process of the sleeve portion inserted into the induction coil, using at least one or more different characteristic quantities as described in claim 1.

9. The induction coil unit according to claim 8, wherein the self-learning system determines the monitoring characteristic quantity using at least one further characteristic quantity as described in claim 2.

10. The induction coil unit according to claim 8 or 9, characterized by at least one current detector and / or voltage detector in an intermediate circuit of a frequency converter that supplies electrical energy to a coil circuit, and / or in a coil circuit having the induction coil.

11. An induction coil unit according to any one of claims 8 to 10, comprising a comparator for comparing at least one characteristic quantity, such as a monitored characteristic quantity and / or a characteristic quantity supplied to a self-learning system according to claim 1 and / or claim 2, with a predetermined limit value.

12. An induction coil unit according to any one of claims 8 to 11, characterized by an automatic shut-off device configured to shut off (or at least modify) the heating process by the induction coil when a limit value is reached in comparison, or when the limit value is lowered or exceeded.

13. An induction coil unit according to any one of claims 8 to 12, characterized by at least two self-learning systems, particularly operating in parallel, for determining a plurality of different monitoring characteristics, in particular a temperature characteristic and a sleeve characteristic.

14. A self-learning system, particularly a neural network or decision tree, for monitoring an induction heating process, A self-learning system, particularly a neural network or decision tree, configured to determine monitoring characteristics, such as a temperature characteristic of a sleeve portion inserted into an induction coil, and / or a sleeve characteristic of the sleeve portion inserted into the induction coil, and / or a time characteristic of the heating process of the sleeve portion inserted into the induction coil, using at least one or more different characteristic quantities as described in claim 1.

15. A self-learning system according to claim 14, in particular a neural network or decision tree, configured to determine the monitoring characteristic using at least one further characteristic described in claim 2.

16. A self-learning system, particularly a neural network, according to claim 14 or 15, characterized by multiple hidden layers, particularly 10 or more hidden layers, particularly 15 or more hidden layers, particularly 20 or more hidden layers, particularly 25 or more hidden layers, particularly 30 or more hidden layers, and particularly 32 hidden layers.

17. The self-learning system, particularly the neural network, according to any one of claims 14 to 16, characterized in that a ReLU function or a sigmoid function is used as the activation function in the self-learning system, particularly the neural network.

18. A self-learning system, particularly a decision tree, according to any one of claims 14 to 16, characterized in that multiple decision trees, in particular decision trees having the same structure, are used, and in particular, the selection of the multiple neural networks to be used is based on the criterion of differences between the decision trees.

19. A self-learning system according to any one of claims 14 to 18, in particular a method for training a neural network or a decision tree, A method for training a self-learning system, particularly a neural network or a decision tree, characterized in that, under the use of at least one or more different characteristic quantities as described in claim 1, the self-learning system is trained to determine a monitoring characteristic quantity, such as a temperature characteristic quantity of a sleeve portion inserted into an induction coil, and / or a sleeve characteristic quantity of the sleeve portion inserted into the induction coil, and / or a time characteristic quantity of the heating process of the sleeve portion inserted into the induction coil.

20. A method for training a self-learning system, particularly a neural network or a decision tree, according to claim 19, characterized in that at least one further characteristic described in claim 2 is also used for training.

21. A method for training a self-learning system, particularly a neural network or decision tree, according to claim 19 or 20, wherein monitoring characteristics such as the temperature characteristic quantity and / or the sleeve characteristic quantity and / or the time characteristic quantity are also used in training, wherein the monitoring characteristics used in training are determined under the use of a defined and performed heating process, particularly measured in the heating process, or determined from quantities measured in the heating process.

22. A method for training a self-learning system, particularly a neural network or a decision tree, according to any one of claims 19 to 21, characterized in that the training is performed using a backpropagation method or the CART- and / or Scikit-learn algorithm.

23. A method for training a self-learning system, particularly a neural network or a decision tree, according to any one of claims 19 to 22, characterized in that the self-learning system is retrained with the use of user feedback and / or user interaction.

Images