Process for straightening a blade

Abstract

A process for straightening a blade, preferably a circular blade, by hammering comprises the steps of: acquiring information relating to the deformation of the blade, including the initial planarity data, by means of sensors; constraining the blade in the centre thereof by means of a controlled locking flange, which allows free rotation about the blade axis; executing a predictive model based on artificial intelligence algorithms to estimate the point variation in the deformation of the blade following each strike as a function of the blade characteristics, the central constraint and the parameters of the hammering already performed; and executing an optimisation system to determine the optimal hammering sequence to be applied on the blade in order to obtain a predefined final planarity profile, thereby minimising the number of hammer strikes and maximising the overall effectiveness of the process.

Description

[0001] “PROCESS FOR STRAIGHTENING A BLADE”

[0002] DESCRIPTION

[0003] * * * * *

[0004] Technical field

[0005] The present invention fits into the sector of cutting blades, preferably discs, installable on machinery or tools intended for cutting wood, aluminium and stone.

[0006] In particular, the invention relates to a process and a system for straightening circular blades and automatically restoring the planarity of the cutting disc.

[0007] * * * * *

[0008] Prior art

[0009] Generally, the non-planarity of the disc represents one of the factors that most greatly limit the performance of a circular blade, whether it be new or resharpened following wear. During cutting operations, the blade is subjected to impacts and thermal peaks which can generate multiple plastic and permanent deformations in the disc, as described in the document Yu et al., 2023; doi: 10.1016 / j.compag.2023.108042.

[0010] At present, as described in patent application US4450707, the process of straightening components with low ductility is based on manual hammering of the surface, which induces a localised loosening of the material. Because of the complexity of this process, the operation must be carried out by highly specialised personnel, whose competence is founded on know-how that is informal and difficult to transfer.

[0011] Various automated solutions have been proposed over time, but they have not yet had broad industrial application. Among them, patent document CN114210766A discloses an automatic circular blade straightening process based on the collection of a database of manual hammering subsequently processed by a deep learning algorithm in order to generate a strategy of straightening by means of a robotic manipulator. However, this process requires the intervention of an expert operator for the collection of the initial dataset, thus introducing a bias tied to the implicit know-how of the operator and limiting the degree of automation of the solution.

[0012] The subsequent patent document EP4442408A1 partly addresses this problem by introducing an automated data collection system. However, the method proposed requires the carrying out of specific preliminary experiments, which are potentially destructive, before the actual implementation. Moreover, the process, based on predictive and prescriptive models of deformations and hammer strikes, is applied on components placed on a surface (for example an anvil) without any fastening constraints. The absence of a constraint can compromise the repeatability of the operations, which is a critical aspect for industrial automation (see Kuric et al., 2020; doi: 10.1177 / 1729881420905723), and reduce the effectiveness of the hammer strikes, whose energy is partly dissipated through the vibration of the component.

[0013] The literature confirms that the presence of a fixing system during mechanical operations significantly influences their outcome (Muhoa et al., 2016; doi:

[0014] 10.1016 / j.cirp.2O16.06.004; Zeng et al., 2012; doi: 10.1016 / j.ijmachtools.2012.02.008). More recent studies empirically demonstrate that fixing the components is particularly relevant in the case of elements with low rigidity, such as blades (Haibo et al., 2022; doi: 10.1007 / s11465-022-0711 -5), and that different constraint configurations decisively influence the distribution of mechanical deformations (Li et al., 2023; doi:

[0015] 10.1038 / S41598-023-33666-2; Croppi et al., 2019; doi: 10.3390 / machines7040068).

[0016] * * * * *

[0017] Summary

[0018] In orderto overcome the limitations of the prior art and bring an improvement to existing methods, the present invention proposes an automated straightening system with central fixing of the cutting blade, which makes it possible to reduce the dispersion of the impact force and to improve the repeatability and uniformity of the operations.

[0019] The present invention also introduces a completely automated pipeline which, by means of automatic learning techniques, is capable of reconstructing the shape of the workpiece, defining areas of exclusion, acquiring and interpreting the deformation values, and predicting, based on the fixing configuration and the previously acquired data, the evolution of the deformation resulting from the hammer strikes. The system thus determines an optimal policy capable of reducing the number of strikes necessary, thereby maximising effectiveness at the same time. In particular, one object of the present invention is to be immediately operational and progressively improve its precision at every iteration by virtue of automatic learning mechanisms, thus avoiding the need to involve persons skilled in the art or carry out specific experiments to calibrate the system.

[0020] The stated technics! task and specified objects are substantially achieved by a process for straightening a blade, which comprises the technical features disclosed in the independent claim. The dependent claims correspond to further advantageous aspects of the invention.

[0021] The invention relates to a system (and a corresponding process) for straightening a blade, preferably a circular blade, constrained in the centre and subjected to hammering on both faces thereof.

[0022] In detail, the aforesaid system comprises: - a device for acquiring the deformity of the blade, comprising a sensor capable of detecting the discrepancy between an ideal reference plane and the surface of blade itself. Preferably, the sensor is a probe or another analogue or digital measuring device with sufficient precision to characterise the deformation of the disc; - computer-executable software for calculating a “neutral” plane, i.e. a theoretical reference plane which represents the condition of ideal contact between the blade and the fixing system, determined in such a way as to eliminate the influence of the support and of the locking flange on the measurement of the deformation. The calculation of the neutral plane makes it possible to obtain an absolute measurement of the blade deformation which is precise, repeatable and comparable between different blades, thus assuring the consistency of the data acquired for subsequent processing or comparisons;

[0023] - a support for the blade configured to constrain the blade in the centre thereof by means of a locking flange capable of applying a pre-established, controllable force to hold the blade itself in position, while leaving the latter free to rotate about its normal rotation axis. The support allows access to both faces of the blade for the execution of the hammer strikes, assuring a controlled, symmetrical deformation during the process;

[0024] - an actuator provided with a hammering tool consisting of a device, preferably a robot, equipped with a tool capable of striking the surface of the blade at least once on each face. The actuator is controlled by pre-defined parameters, including: o the distance from the centre of the blade, measured along the radius of the disc; o the angle of incidence relative to the initial position of the blade on the support; o the force to be applied and the direction of the strike, indicating the face of the blade to be struck; - analytical software configured to identify the shape of the blade based on a technical representation thereof, preferably two-dimensional, understood as a digital model obtained from a CAD drawing or an equivalent technical file containing the geometric information of the component (i.e. the blade on which to perform machining), and to identify the areas of exclusion to be considered during the selection of the hammering actions, such as, for example, holes, slots or teeth of the blade. The analytical software is configured to receive, as input, a digital technical drawing of the blade and to generate, as output, a binary mask in polar coordinates indicating the areas of exclusion, and subsequently used by the optimisation system to generate the hammering policy; - a predictive model configured to estimate the variation in the deformation of the blade following the straightening strikes. The predictive model is configured to receive as input: o the technical specifications of the blade (hardness, inner radius, outer radius); o the characteristics of the locking flange (diameter, applied force); o the mask obtained from the analytical software; o the initial reading of the distortion; o an empirically derived physical model of deformation of the blade; o the number, position, force, direction and sequence of the strikes already applied; o the parameters of the straightening strike it is intended to apply (distance from the centre, angle of incidence, force, direction).

[0025] The predictive model is further configured to return, as output, the precise predicted variation of the distortion resulting from the strike. It is constructed through artificial intelligence or other mathematical modelling techniques, preferably a model-based artificial neural network (MBANN), and is continuously trained based on the historical data collected by the system (blade and flange characteristics, distortion readings and parameters of the hammer strikes performed); - an optimisation system configured to identify a hammering strategy that minimises the number of strikes and maximises the overali effectiveness of the process. Supported by the predictive model described previously, the system computes the best sequence of actions the actuator must carry out in order that the point deformation of the blade remains below a threshold predefined by the user. The system is based on an artificial intelligence model or a mathematical / analytical approach, preferably a generative pretrained transformer (GPT) or equivalent architecture, and receives as input: o a point reading of the deformation; o the blade characteristics; o the exclusion mask.

[0026] As output, the system generates an optimal sequence of actions the actuator must carry out to reach the desired planarity.

[0027] The analytical software, the predictive model and the optimisation system are operations performed by a computer or processor or microcontroller or similar components of a digital processing system.

[0028] In other words, the aforesaid system is advantageously capable of establishing a blade planarization strategy based on the blade’s geometric characteristics and deformities, and, moreover, as a function of the planarity modelled predictively with the aid of a suitably trained artificial intelligence or mathematical / analytical model, preferably an MBANN. The optimisation system, in fact, is advantageously capable of selecting a sequence of hammer strikes applicable on the blade to be straightened with the aim of obtaining a predefined final planarity profile within the various tolerance limits so that the blade is planar and thus substantially free of surface defects. Therefore, the implementation of the system makes it possible to create a straightening apparatus capable of operating autonomously in performing measurements, processing data and, above all, elaborating the operational strategy with which to obtain the straightening of a blade, without any need for external inputs from operators, which are sometimes subjective and not replicable. Furthermore, the present invention advantageously takes account of the geometric variations typical of different cutting blades, which can negatively impact the effectiveness of the straightening strikes. The analytical software integrated into the system is configured to autonomously recognise such variations and identify the areas to be excluded during the straightening process in order to prevent undesirable deformations of the component. In detail, the software receives, as input, a geometric representation of the blade, preferably two-dimensional, and, through the recognition of recurrent schemes in the geometries, such as silencing slots, scrapers or teeth, automatically distinguishes the openings or holes from the solid surface of the disc. The areas of exclusion are thus defined as the smallest rectangular bounding box that circumscribes each hole or opening detected, providing the optimisation system with a precise reference for the generation of the sequence of hammer strikes.

[0029] The same areas of exclusion are also used during the phase of training of the optimisation system, which integrates them into its decision model, automatically excluding them from the generation of the hammering policy.

[0030] By virtue of the configuration of the blade support, the invention makes it possible to operate on both faces of the component without any need for a manual intervention, thus reducing the risks for the operator and enabling the execution of the process irrespective of the orientation of the blade inside the apparatus. The central constraint of the blade reduces the dissipation of the force applied by the actuator, thus enabling a more efficient transmission of the impact energy and a greater uniformity of the treatment.

[0031] In accordance with a preferred aspect of the invention, this configuration favours a more homogeneous radial distribution of the deformation, thus increasing the effectiveness of the process in terms of symmetrical geometries and improving the overall quality of the planarisation. Furthermore, the system comprises a software-implemented procedure for calculating the so-called “neutral plane” of contact between the disc and the fixing support, an essential step for the correct measurement of the deformation of the component. The calibration of the entire system, supported by the software procedure, enables absolute and repeatable measurements to be obtained, eliminating the influence of the fixing system. Since every measurement depends on the relative position between the object and the transducer, the supports used can also influence the acquired data; the calibration and the calculation of the neutral plane make it possible to remove this influence and to define a common reference, thereby assuring a consistent measurement of the deformation, comparable between different components and usable for subsequent processing or comparisons.

[0032] In accordance with a preferred aspect of the invention, in the event that it is necessary 5 to perform a plurality of hammer strikes on the blade, the steps of the process (in particular the steps of executing the predictive model and the strategic algorithm) may be iterated a number of times with the aim of improving the precision and accuracy of the hypothetical planarity profile and, consequently, with the aim of further optimising the hammering strategy derivable from the strategic algorithm.

[0033] 10 In general, the aforesaid process is advantageously capable of implementing a greater control over the flow of the blade machining steps. In this manner, it is advantageously possible to trace and certify the entire process, thereby assuring a greater quality of the final product (compared to the same products obtainable with the machining processes of the prior art).

[0034] -1i. r; * * * * *

[0035] Brief description of the drawings

[0036] Additional features and advantages of the present invention will emerge more clearly from the approximate and thus non-limiting description of a preferred but not exclusive embodiment of a process for straightening a blade, as illustrated in the appended 20 drawings, in which: figure 1 illustrates, according to a schematic view, a flow diagram representative of a preferred embodiment of the process for predicting the optimal straightening sequence; figures 2 and 3 show two examples of a comparison between the effect deriving

[0037] 25 from the application of hammer strikes in angular positions close to each other; figure 4 shows a graph comparing a predicted planarity profile and a planarity profile as actually measured (following the application of predicted optimal hammer strikes); figures 5 and 6 illustrate two examples in which a planarity profile predicted by

[0038] 30 the artificial intelligence algorithm and an actually measured planarity profile are compared for a same blade; figure 7 illustrates a schematic representation of the fixing block, including the locking flange, which allows access to both faces of the blade.

[0039] With reference to the drawings, they serve solely to illustrate embodiments of the invention with the aim of better clarifying, in combination with the description, the inventive principles at the basis of the invention.

[0040] Detailed description of at least one embodiment

[0041] Figure 1 shows a preferred embodiment of a process for straightening a blade, preferably a circular blade, by means of corrective actions such as hammer strikes performed with a corresponding straightening system as described below. The process for straightening a blade 100, preferably a circular blade, comprises the steps of:

[0042] - determining a reference plane for measuring the deformation of the blade to be straightened 100, suitable for providing absolute values that are repeatable and comparable between different blades; ~ constraining said blade to be straightened 100 on a support 110 configured to allow access to both surfaces of said blade 100 to be straightened 100 in order to ensure a controlled, symmetrical deformation;

[0043] - applying one or more corrective actions on at least one surface of said blade to be straightened 100, each corrective action being defined by position, intensity, direction and sequence parameters;

[0044] - identifying any surface areas of said blade to be straightened 100 in order to exclude them from said corrective actions, said identification being based on a technical representation of the blade to be straightened 100, preferably two- dimensional; - predicting the variation in the deformation of the blade to be straightened 100 resulting from each corrective action by means of a predictive model configured to receive, as input, the characteristics of said blade to be straightened 100 and of said support 110, the measured deformation, the history of the actions applied and the parameters of the actions provided for; - optimising the sequence of corrective actions based on the results of the predictive model, with the aim of minimising the number of corrective actions necessary in order to ensure that the deformation remains within pre- established limits.

[0045] Preferably, the process can comprise preliminary steps of acquiring a first and a second dataset. The first dataset is representative of information relating to the blade to be straightened, including a diameter value of the blade, a thickness value of the blade, a diameter value of a blade locking flange, and a planarity profile of the blade before and after the performance each hammering. The second dataset, by contrast, is representative of information relating to the hammering performed on the blade to be straightened, including, for each hammer strike performed, an intensity value of the hammer strike, a radial hammering position relative to the centre of the blade, and a value of an angular hammering position relative to a pre-established reference point on the surface of the blade.

[0046] Below is a description of a possible and preferred implementation of the components used in the abovementioned system and straightening process to identify the optimal straightening sequence: - analytical software configured to identify the shape of the blade based on a two- dimensional technical representation thereof, and to identify the areas of exclusion to be considered during the selection of hammering actions, such as, for example, holes, slots or teeth of the blade. The software receives a digital technical drawing of the blade as input and generates, as output, a binary mask in polar coordinates indicating the areas of exclusion and subsequently used by the optimisation system for the generation of the hammering policy:

[0047] - a predictive model, intended to estimate the variation in the deformation of the blade following the straightening strikes. The model receives as input: o the technical specifications of the blade (hardness, inner radius, outer radius); o the characteristics of the locking flange (diameter, applied force); o the mask obtained from the analytical software; o the initial reading of the distortion; o an empirically derived physical model of deformation of the blade; o the number, position, force, direction and sequence of the strikes already applied; o the parameters of the straightening strike it is intended to perform (distance from the centre, angle of incidence, force, direction). The model returns, as output, the predicted point variation in the distortion resulting from the hammer strike. It is based on a model-based artificial neural network (MBANN) architecture, and is trained continuously based on the historical data collected by the system (blade and flange characteristics, distortion readings and parameters of the hammer strikes performed);

[0048] - an optimisation system, configured to identify a hammering strategy that minimises the number of strikes and maximises the overall effectiveness of the process. Supported by the predictive model described previously, the system computes the best sequence of actions the actuator must carry out in order that the point deformation of the blade remains below a threshold predefined by the user. The system is based on a generative pretrained transformer (GPT), and receives as input: o a point reading of the deformation; o the blade characteristics; o the exclusion mask.

[0049] As output, the system generates an optimal sequence of actions the actuator must carry out to reach the desired planarity.

[0050] The analytical software, the predictive model and the optimisation system are operations performed by a computer. Advantageously, therefore, the aforesaid process is capable of optimising the hammering of a circular cutting blade by identifying what the best parameters for hammering are (one hammer strike or a succession of hammer strikes). The process is then capable of producing a report which describes an efficiency-enhanced hammering strategy that can be implemented by a specific blade straightening system following the entry, direct or manual, of the aforesaid optimised hammering parameters

[0051] (i.e. the intensity values of each hammer strike and the respective angular and radial positions).

[0052] A three-dimensional planarity profile of the blade is created with the combination of the measurements of the planarity profile for each diameter of the same blade. From this, one notes that the deformation of the blade is nearly linear along a radius thereof.

[0053] Therefore, the planarity profile of a blade depends mainly on the diameter of the locking flange and, moreover, on the closing force of the locking flange which, for the sake of simplicity, may be assumed as a constant value for every straightening machine.

[0054] The step of executing a predictive model aims to describe the mechanical effect of each hammer strike in order to be able to determine, in a subsequent step, the optimal hammer strikes to be performed to planarise the same blade, in accordance with the pre-established planarity tolerance limits.

[0055] In other words, the predictive model makes it possible to predict, for each point of the planarity profile of the blade, the surface plastic deformation caused by a hammer strike having specific characteristics in order to create a hypothetic planarity profile of the blade resulting from the aforesaid hammer strike and which can subsequently be used by the strategic algorithm to create a planarisation strategy.

[0056] In accordance with one aspect of the invention, the step of executing a predictive model comprises integrating a physical model of the blade deformation, obtained empirically in a laboratory, with an artificial neural network.

[0057] The physical model allows for reproducing the mechanical behaviour of the workpiece in response to the straightening strike, whilst the neural network is configured to consider further factors not included in the general physical model, such as, for example, the presence of holes, teeth or geometric discontinuities in the blade.

[0058] The output of the neural network is used to correct the result of the physical model, making it possible to obtain a prediction that is more accurate and in line with the real plastic behaviour of the component.

[0059] In the case of hammer strikes repeated in the same angular position of the blade (or in different angular points that may be grouped together in a same angular interval with a width of about 10°), for every hammer strike after the first, the corresponding surface modelling decreases significantly due to a different mechanical response of the blade (in particular due to the metal material making up the blade, for example steel).

[0060] In accordance with one aspect of the invention, therefore, during the step of executing a predictive model for every point of the planarity profile, a check is performed regarding the angular (and radial) position of the hammer strikes that took place previously. For example, if one or more hammer strikes have been performed in an interval of about 10° relative to the angular position of the first hammer strike taken into consideration, a reduction of the effect of the hammer strike on the blade is estimated. The graphs in figures 2 and 3 show two examples of a comparison between the effect deriving from the application of hammer strikes in angular positions close to one another.

[0061] In detail, the graphs illustrate the ratio, in percentage terms, between the shift of the point of hammer strike and the planarity in that point before the hammer strike. For example, a value equal to 30 indicates that, if the point was at 100 pm relative to the ideal reference of planarity, the shift due to the hammer strike will be 30 pm. On observing the aforesaid graphs, one notes that the effect of the hammer strike repeated in the same area of the blade tends to decrease significantly for subsequent hammer strikes and can be approximated as 30% of the original effect.

[0062] The physical model of the blade used in the predictive model takes account of the radial propagation of the force and the decrease in the plastic response due to the repetition of strikes in the same surface area. In particular, the variation of the deformation Ad(6,r) in one point of the blade, defined by the angle 0 and the radius r, may be expressed as:

[0063] Ad(6,r) ~ F0fang (& - Oo)0frad (fjo)0fsat (H(9j)) where:

[0064] « F is the intensity of the strike applied;

[0065] * fang (3 ~ 3o) ~ exp (- (9 - 3o)2 / 2o2) represents the angular decrease in the force around the strike angle 3o, with a tied to the extent of angular propagation:

[0066] » fraa (r,ro) = exp (~ kcro - kd represents the radial decrease in the plastic response as a function of the distance from the centre of the strike ro and the radius of size r, with the coefficients kc, kd empirically determined;

[0067] * fsat (H(3,r)) = exp @ H(3,r)) is the saturation factor that reduces the effect of repeated strikes as a function of the cumulative action H(3,r) already applied on the point;

[0068] « H(3,r) represents the memory of the previous strikes applied to the same area of the blade.

[0069] This formulation makes it possible to estimate, in a physically coherent manner, the point variation in the deformation of the blade, before any correction learnt by the neural model.

[0070] The neural network associated with the predictive model is trained on the data collected during the automatic straightening process, stored in a historical database that correlates the planarity readings to the hammer strikes performed and to the respective sequences. In this manner, the model progressively learns the effect of each hammer strike as a function of the initial conditions of the workpiece. Advantageously, the model is capable of predicting the elastic and plastic response of the blade also in the case of shapes or configurations not previously observed, by encoding the input parameters in a reduced-order latent space, in which a search is made for the closest action among those present in the historical database.

[0071] This approach enables the model to be used right from the start without the need for dedicated pre-training, allowing the system to improve its performance progressively with the incremental use and continuous collection of new operating data.

[0072] In relation to what was just described, figures 5 and 6 show two examples in which a comparison is made between the values relating to the hypothetical planarity profile of the blade resulting from the execution of the predictive model (dashed line) and the values relating to the planarity profile of the blade actually measured following the application of the hammer strike (solid line). Two reference lines are traced at ±10 pm relative to the measured profile in order to provide a further visual comparison of the error deriving from the predictive model.

[0073] The step of executing an optimisation system, as previously described, uses the predicted output values resulting from the execution of the predictive model (i.e. the hypothetical planarity profile of the blade) in order to identify a succession of optimal, i.e. effective, hammer strikes with which it is possible to obtain a pre-established final planarity profile of the blade that is consistent with the tolerance limits decided beforehand and is moreover possibly obtained with the smallest number of hammer strikes.

[0074] In accordance with a preferred aspect of the invention, the optimisation system comprises a model of the generative pretrained transformer (GPT) type, configured to associate the blade planarity reading with the optimal straightening action necessary to bring the subsequent state within a target value selected by the user. The model processes the planarity reading, projecting it into a higher-order latent space, where it is concatenated with the physical parameters of the blade (inner diameter, outer diameter and hardness), the parameters of the locking flange (diameter and applied force) and the exclusion mask generated by the analytical software.

[0075] Each configuration thus encoded is associated with the corresponding hammering action so that the model, during subsequent iterations, is capable of recognising similar states and automatically proposing the most effective action. The advantage of this approach lies in the ability of the model to identify the optimal hammering policy irrespective of the characteristics of the blade entered into the system.

[0076] By virtue of its architecture, every new state is projected into the higher-order space, where the closest configuration is searched for; the action associated with that configuration is then selected and suggested for the next step, thus enabling continuous learning and adaptation of the system.

[0077] In accordance with a preferred aspect of the invention illustrated in figure 1 , the process also comprises a step of repeating the steps of the entire straightening process, in particular at least the steps of executing a predictive model and a strategic algorithm, with the aim of optimising the hammering sequence by minimising the number of hammer strikes to be performed in order to obtain the aforesaid predefined final planarity profile of the blade.

[0078] It is possible, in fact, that it will be necessary to apply a plurality of hammer strikes on the same blade in order to obtain a planarity profile that complies with the tolerance limits established beforehand.

[0079] Consequently, once the information relating to the first hammer strike to be applied to the blade has been established and the resulting corresponding hypothetical planarity profile has been calculated (and the respective first and second datasets have been acquired), the latter can be used as a new initial planarity profile to which the steps of the straightening process in accordance with the present invention can again be applied.

[0080] In this manner, therefore, it is advantageously possible to obtain information relating to the second hammer strike to be applied to the blade and, moreover, the corresponding hypothetical planarity profile of the blade. Advantageously, the number of iterations of the straightening process can be extended until the tolerance limits established beforehand are met and / or a predetermined maximum number of iterations is reached. In fact, the error in the prediction of the final planarity profile of the blade grows significantly as the number of iterations carried out increases.

[0081] Figure 4 shows a graphic representation of the planarity profile predicted on a test blade on which a sequence of optimal hammer strikes identified by the process of the invention was applied. The prediction is then compared with the actual planarity profile resulting from the application of the sequence of hammer strikes performed using manually selected operating parameters.

[0082] In detail, figure 4 compares the planarity profile measured following manual hammer strikes (dotted line) with the predicted one (dashed line) relative to the performance of the optimal hammer strikes.

[0083] In reality, 8 iterations of the straightening process were carried out; thus 8 hammer strikes of the same intensity were applied to the blade.

[0084] However, the straightening process determined that the tolerance limit provided for can be reached with the application of only 4 optimised hammer strikes, thus considerably improving performance in terms of both the quantity of hammer strikes and the quality of the final result. The process thus shows to be advantageously capable of enhancing the efficiency of the blade straightening process by minimising the number of hammer strikes to be applied and, consequently, improving the quality of the final product. In accordance with a further aspect of the invention, between each step of executing a predictive model and each step of executing a strategic algorithm it is possible to apply a real hammer strike on the blade in a radial and / or angular position and with an intensity value predicted by the predictive model.

[0085] In addition, following the real hammer strike, it is possible to determine the corresponding real planarity profile comparable with the aforesaid hypothetical planarity profile in order to provide said strategic algorithm with real input values rather than hypothetical ones so as to further increase the efficiency in deriving the subsequent hammering strategy from the same strategic algorithm and also increase the quality and precision of the final planarity profile and thus of the final product. The present invention also relates to a system for straightening a blade, preferably a circular blade, which comprises: a work surface, for example a mandrel, on which a blade to be straightened can be constrained. In particular, the work surface comprises a motorised locking flange in order to allow the rotation of the blade about its normal rotation axis; a tool configured to hammer the blade in different radial and / or angular positions relative to the blade for the purpose of straightening it; - at least one transducer configured to acquire one or more items of information relating to said blade and / or said hammer strike applicable to said blade by said tool in order to define a first dataset; a control unit associated with the tool and with each transducer and further configured to apply a straightening process comprising one or more of the previously described features.

[0086] The straightening system is thus advantageously capable of automating the previously described straightening process, since the operator’s job is reduced to positioning the blade on the work surface and coupling it with the locking flange accordingly.

[0087] Even more advantageously, therefore, the system is capable of making the operator’s job less arduous, thus improving the quality of the work thereof, increasing productivity and, above all, implementing a replicable hammering strategy based on objective information.

[0088] In accordance with a further aspect of the invention illustrated in figure 7, the central fixed constraint of the blade 100, achieved by means of a locking flange 111 , 112, plays a fundamental technical role for the purposes of the effectiveness and repeatability of the straightening process.

[0089] Unlike the known systems in which the blade is simply rested on a flat surface or partially constrained, the presence of a stable central fixing point makes it possible to eliminate undesirable degrees of freedom during the impact of the hammer strike, thereby reducing the dissipation of the applied force and improving the consistency of the elastoplastic response of the material.

[0090] This configuration makes it possible to obtain a symmetrical radial behaviour of the deformation, thus facilitating a uniform propagation of the impact wave from the point of application towards the outside of the blade. The fixed constraint further makes it possible to define an absolute reference system for the acquisition of planarity readings and the encoding of geometric data in the predictive model, ensuring that each strike is interpreted in a clear and repeatable manner by the control system.

[0091] In terms of automatic learning, the presence of the central constraint represents a further advantage: since the conditions around the workpiece remain constant between one cycle and another, the data collected are statistically consistent, allowing for a more rapid convergence of the predictive model and a greater precision in estimating the real response of the blade to the subsequent strikes.

[0092] Furthermore, the fixed constraint reduces the variability due to positioning errors or slipping of the workpiece, thus improving the quality of the dataset used for training and increasing the ability of the system to learn generalisable hammering policies. In accordance with a further aspect of the invention illustrated in figure 7, the work surface is configured in such a way as to enable a bilateral application of the hammering strikes, i.e. the tool is adapted to act selectively on both sides of the blade (as schematically represented by the four arrows present in the figure).

[0093] This configuration is particularly advantageous, as it makes it possible to compensate for any residual deformations or internal tensions that are generated as a result of repeated hammer strikes on one side only, thus improving the overall effectiveness of the straightening process.

[0094] The system can comprise two opposing tools, or alternatively, a single mobile tool driven by linear actuators, capable of positioning itself automatically on the upper or lower side of the blade based on the computed straightening strategy.

[0095] Furthermore, the transducers or other analogous components such as sensors and detectors can be advantageously configured to acquire one or more of the items of information used for the models, send them to the control unit for the storage thereof and formation of a historical database and, subsequently, for the application - possibly iterated - of the steps of the straightening process in order to identify an efficient hammering strategy consisting of one or more optimised hammer strikes (i.e. whose parameters have been suitably calculated to obtain the maximum straightening performance). Preferably, in addition, the system is also configured to minimise the number of hammer strikes to be applied in order to obtain a straightened blade of better quality and within the initially established tolerance limits.

Claims

CLAIMS1. A process for straightening a biade (100), preferably a circular blade, comprising the steps of:- determining a reference plane for measuring the deformation of the blade to be straightened (100), suitable for providing absolute values that are repeatable and comparable between different blades;- constraining said blade to be straightened (100) on a support (110) configured to allow access to both surfaces of said blade (100) to be straightened (100) in order ensure a controlled, symmetrical deformation;- applying one or more corrective actions on at least one surface of said blade to be straightened (100), each corrective action being defined by position, intensity, direction and sequence parameters;- identifying any surface areas of said blade to be straightened (100) in order to exclude them from said corrective actions, said identification being based on a technical representation of the blade to be straightened (100), preferably two- dimensional;- predicting the variation in the deformation of the blade to be straightened (100) resulting from each corrective action by means of a predictive model configured to receive, as input, the characteristics of said blade to be straightened (100) and of said support (110), the measured deformation, the history of the actions applied and the parameters of the actions provided for;- optimising the sequence of corrective actions based on the results of the predictive model, with the aim of minimising the number of corrective actions necessary in order to ensure that the deformation remains within pre- established limits.

2. The process according to claim 1 , wherein the corrective actions consist in hammering or mechanical applications localised on the surface of the blade.

3. The process according to claim 1 or 2, comprising the use of one or more sensors configured to detect the deformation of the blade to be straightened (100) relative to said reference plane with sufficient precision in order to characterise point variations of the surface.

4. The process according to any one of the preceding claims, wherein the surfaceareas to be excluded are identified by means of analytical software that receives, as input, a technicai representation of the blade and generates, as output, exciusion masks, preferably in polar coordinates, for the subsequent optimisation of the corrective actions.

5. The process according to any one of the preceding claims, wherein the predictive model combines a physical model of the blade deformation with automatic learning techniques or artificial intelligence, for the purpose of estimating the point variation in deformation resulting from the corrective actions.

6. The process according to any one of the preceding claims, comprising a step of repeating at least the prediction and optimisation steps, for the purpose of optimising said succession of corrective actions, thereby minimising the number of corrective actions necessary to obtain the predefined final planarity profile of the blade to be straightened (100), taking account of a central constraint applied by a locking flange and the free rotation of the blade to be straightened (100) about a rotation axis thereof7, The process according to any one of the preceding claims, wherein said step of executing a predictive model, for each value of an angular hammering position, is configured to maintain said second dataset unchanged and translate the planarity profile of the blade before the hammering in relation to the angular position of the hypothetical hammer strike.8, The process according to any one of the preceding claims, comprising a step of training a neural network associated with said predictive model in order to calculate the variance in planarity following the performance of a straightening strike.

9. The process according to any one of the preceding claims, wherein said optimisation step comprises executing a strategic algorithm based on a generative pretrained transformer to encode the inputs and identify the optimal actions.10, A system for straightening a blade (100), preferably a circular blade, comprising: a work surface, for example a mandrel, on which a blade to be straightened can be constrained, said work surface comprising a locking flange (111 , 112) for locking the blade to be straightened (100), configured to allow the rotation of the blade to be straightened (100) about the normal rotation axis thereof; a tool configured to hammer the blade with a variable intensity in different radial and / or angular positions relative to the blade for the purpose of straightening it;at feast one transducer configured to acquire one or more items of information relating to said blade to be straightened (100) and / or to said corrective action applicable to said blade by said tool in order to define a first and / or a second dataset, wherein said first dataset is representative of information relating to the blade to be straightened (100), including a diameter value of the blade, a thickness value of the blade, a diameter value of a blade locking flange, and a planarity profile of the blade before and after the performance of each hammering, wherein said second dataset is representative of information relating to the hammering performed on the blade to be straightened, including, for each hammer strike performed, an intensity value of the hammer strike, a radial hammering position relative to the centre of the blade, and a value of an angular hammering position relative to a pre-established reference point on the surface of the blade; a control unit associated with said tool and said at least one transducer and configured to apply a straightening process in accordance with any one of claims 1 to 9.

Images