System and method for estimating reliability of thermal error prediction model for a machine tool
Patent Information
- Application Number
- IN202541051858
- Authority / Receiving Office
- IN · IN
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2025-05-29
- Publication Date
- 2026-08-04
- Estimated Expiration
- 2045-05-29
Abstract
Description
TECHNICAL FIELD
[0001] The embodiments of the present disclosure generally relate to the field ofmachine tool condition monitoring and performance assessment. More particularly, thepresent disclosure relates to a system and a method for estimating a reliability of amodel used for predicting thermal error of a machine tool.BACKGROUND OF THE INVENTION
[0002] The subject matter disclosed in the background section should not be assumedor construed to be prior art merely because of its mention in the background section.Similarly, any problem statement mentioned in the background section or itsassociation with the subject matter of the background section should not be assumedor construed to have been previously recognized in the prior art.
[0003] In modern manufacturing industries, continuous production of high-precisioncomponents is essential to meet a growing demand for quality and efficiency. Machinetools are responsible for an accurate machining of components. However, performanceof the machine tools is significantly affected by various sources of error, among whichthermal error is recognized as one of the most dominant contributors. Thermal errors,induced by changes in temperature during machine operation, lead to dimensionaldeviations, ultimately affecting an accuracy and quality of a machined product.
[0004] Heretofore, numerous thermal error compensation strategies have beenproposed to mitigate adverse effects of thermal variations on machine toolsperformance. Among these, models used for predicting thermal errors are gaining wideacceptance due to their cost-effectiveness and ease of integration into existing systems.By leveraging data-driven or physics-based approaches, these models predict thermalinduceddeviations, allowing real-time compensation to be applied during machiningoperations. However, effectiveness of these models is highly dependent on theirrobustness and reliability, particularly in dynamic production environments wheremachine conditions are continuously evolving.
[0005] Challenges faced by existing thermal error compensation approaches inmaintaining the accuracy of the models used for predicting the thermal errors includedegradation of the machine tools and temperature sensors over time. Factors such asmechanical wear, component aging, and sensor drift introduce uncertainties into aprediction process, leads to a gradual decline in model performance. Despite a criticalimpact of these factors, the existing thermal error compensation approaches largely relyon conventional error metrics, such as Root Mean Square Error (RMSE), MeanAbsolute Error (MAE), and Mean Absolute Percentage Error (MAPE). These metrics,while useful for initial model evaluation, do not adequately capture reliability of thesemodels under varying operating conditions or account for sensor degradation effects.Thus, these approaches are often ad-hoc, lacking a systematic framework for real-timemonitoring and reliability assessment of these models during deployment.
[0006] Therefore, to overcome aforementioned challenges and limitations associatedwith the conventional systems, there lies a need for an improved system and a methodfor estimating the reliability of the models used for predicting the thermal errors of themachine tools.OBJECTS OF THE INVENTION
[0007] An object of the present disclosure is to provide a system and a method toaccurately estimate a reliability of a deployed model used for predicting thermal errorsby analyzing effects of sensor degradation and machine tool condition variations onmodel performance.
[0008] Another object of the present disclosure is to provide a system and a methodthat facilitates real-time data acquisition from machine tool hotspots, seamlesscommunication of compensation values to machine controllers, and use cloud-basedplatforms for model reliability assessment, decision-making, and maintenanceplanning.
[0009] Yet another object of the present disclosure is to provide a system and a methodthat enables a periodic evaluation and upgrading of the model used for predicting thethermal errors based on observed degradation of machine tools and sensors.SUMMARY
[0010] The following embodiments present a simplified summary to provide a basicunderstanding of some aspects of the disclosed invention. This summary is not anextensive overview, and it is not intended to identify key / critical elements or todelineate the scope thereof. Its sole purpose is to present some concepts in a simplifiedform as a prelude to the more detailed description that is presented later.
[0011] According to an aspect of the present disclosure, disclosed herein is a methodfor estimating a reliability of a model used for predicting thermal error of a machinetool. The method comprises receiving, from one or more temperature sensors,temperature data associated with one or more positions on the machine tool andmeasuring, based on the received temperature data, a thermal error value at a tool centerpoint of the machine tool. The method further comprises training the model using thereceived temperature data and the measured thermal error value and generatingdegraded temperature data by applying predefined variations to the receivedtemperature data to indicate deterioration in the one or more temperature sensors. Themethod further comprises providing the degraded temperature data to the trained modelto obtain a predicted thermal error value and determining performance metrics databased on a difference between the predicted thermal error value and the measuredthermal error value. The method further comprises estimating the reliability of themodel based on the determined performance metrics data.
[0012] In one or more implementations, the method further comprises determiningwhether a value associated with the performance metrics data falls below a thresholdvalue and determining an information loss based on the determination that the valueassociated with the performance metrics data falls below the threshold value. Themethod further comprises updating the model based on the determination of theinformation loss.
[0013] In one or more implementations, the method further comprises re-estimating,upon the updating of the model, the reliability of the model until the estimatedreliability meets or exceeds a predefined reliability threshold.
[0014] In one or more implementations, the predefined variations include simulationof a noise or a predefined failure criterion in the received temperature data.
[0015] In one or more implementations, the performance metric data comprises valuesof functions selected from one or more of Root Mean Square Error (RMSE) and MeanAbsolute Error (MEA).
[0016] According to another aspect of the present disclosure, a system for estimatinga reliability of a model used for predicting thermal error of a machine tool is disclosed.The system comprises a server including at least one processor. The at least oneprocessor is configured to receive, from one or more temperature sensors, temperaturedata associated with one or more positions on the machine tool. The at least oneprocessor is configured to measure, based on the received temperature data, a thermalerror value at a tool center point of the machine tool. The at least one processor isfurther configured to train the model using the received temperature data and themeasured thermal error value and generate degraded temperature data by applyingpredefined variations to the received temperature data to indicate deterioration in theone or more temperature sensors. The at least one processor is configured to providethe degraded temperature data to the trained model to obtain a predicted thermal errorvalue and determine performance metrics data based on a difference between thepredicted thermal error value and the measured thermal error value. The at least oneprocessor is configured to estimate the reliability of the model based on the determinedperformance metrics data.BRIEF DESCRIPTION OF DRAWINGS
[0017] Various embodiments disclosed herein will become better understood from thefollowing detailed description when read with the accompanying drawings. Theaccompanying drawings constitute a part of the present disclosure and illustrate certainnon-limiting embodiments of inventive concepts. Further, components and elementsshown in the drawings are not necessarily to scale, emphasis instead being placed uponclearly illustrating the principles of the present disclosure. For consistency and ease ofunderstanding, similar components and elements are annotated by reference numeralsin the exemplary drawings.
[0018] FIG. 1 illustrates a block diagram of a system for real-time thermal errorprediction and reliability assessment in machine tools, in accordance with anembodiment of the present disclosure.
[0019] FIG. 2 illustrates a block diagram of an Industrial Internet of Thing (IIoT)module of the system, in accordance with an embodiment of the present disclosure.
[0020] FIG. 3 illustrates a diagram of a circuit board of the IIoT module, in accordancewith an embodiment of the present disclosure, in accordance with an embodiment ofthe present disclosure.
[0021] FIG. 4A illustrates a schematic diagram of thermal sensor locations on amachine tool, in accordance with an embodiment of the present disclosure.
[0022] FIG. 4B illustrates an experimental setup implemented for measuring thermaldrift at a Tool Centre Point (TCP) caused by temperature variations during a machineoperation, in accordance with an embodiment of the present disclosure.
[0023] FIG. 5 illustrates a method for estimating a reliability of a model used forpredicting thermal error, in accordance with an embodiment of the present disclosure.
[0024] FIG. 6 illustrates a detailed flowchart depicting the method for estimating thereliability of a Thermal Error Prediction Model (TEPM) and the upgradation of theTEPM using continuous machine learning framework, in accordance with anembodiment of the present disclosure.
[0025] FIG. 7A illustrates a variation in sensor data under different degradation statesin accordance with an embodiment of the present disclosure.
[0026] FIG. 7B illustrates a comparison of predicted thermal error using differentmachine learning models, in accordance with an embodiment of the present disclosure.
[0027] FIG. 8A illustrates the reliability of prediction models under sensordegradation, in accordance with an embodiment of the present disclosure.
[0028] FIG. 8B illustrates probability distribution of prediction error under the sensordegradation, in accordance with an embodiment of the present disclosure.
[0029] FIG. 9A illustrates prediction error analysis across sensor degradation states,in accordance with an embodiment of the present disclosure.
[0030] FIG. 9B illustrates an exemplary Kullback-Leibler (KL) divergence analysisacross the sensor degradation states, in accordance with an embodiment of the presentdisclosure.
[0031] FIG. 10 illustrates an effectiveness of model upgradation in degraded sensorenvironment, in accordance with an embodiment of the present disclosure.DETAILED DESCRIPTION OF THE INVENTION
[0032] Inventive concepts of the present disclosure will now be described more fullyhereinafter with reference to the accompanying drawings, in which examples of one ormore embodiments of inventive concepts are shown. Inventive concepts may, however,be embodied in different forms and should not be construed as limited to theembodiments set forth herein. Further, the one or more embodiments disclosed hereinare provided to describe the inventive concept thoroughly and completely, and to fullyconvey the scope of each of the present inventive concepts to those skilled in the art.Furthermore, it should be noted that the embodiments disclosed herein are not mutuallyexclusive concepts. Accordingly, one or more components from one embodiment maybe tacitly assumed to be present or used in any other embodiment.
[0033] The following description presents various embodiments of the presentdisclosure. The embodiments disclosed herein are presented as teaching examples andare not to be construed as limiting the scope of the present disclosure. The presentdisclosure should in no way be limited to the illustrative implementations, drawings,and techniques illustrated below, including the exemplary design and implementationillustrated and described herein, but may be modified, omitted, or expanded uponwithout departing from the scope of the present disclosure.
[0034] The following description contains specific information pertaining toembodiments in the present disclosure. The detailed description uses the phrases "insome embodiments" which may each refer to one or more or all of the same or differentembodiments. The term "some" as used herein is defined as "one, or more than one, orall". Accordingly, the terms "one", "more than one", "more than one, but not all" or"all" would all fall under the definition of "some." In view of the same, the terms, forexample, "in an embodiment" refers to one embodiment and the term, for example, "inone or more embodiments" refers to "at least one embodiment, or more than oneembodiment, or all embodiments."
[0035] The term "comprising," when utilized, means "including, but not necessarilylimited to;" it specifically indicates open-ended inclusion in the so-described one ormore listed features, elements in a combination, unless otherwise stated with limitinglanguage. Furthermore, to the extent that the terms "includes," "has," "have,""contains," and other similar words are used in either the detailed description, suchterms are intended to be inclusive in a manner similar to the term "comprising."
[0036] In the following description, for the purposes of explanation, various specificdetails are set forth to provide a thorough understanding of embodiments of the presentdisclosure. It will be apparent, however, that embodiments of the present disclosuremay be practiced without these specific details. Several features described hereaftercan each be used independently of one another or with any combination of otherfeatures.
[0037] The description provided herein discloses exemplary embodiments only and isnot intended to limit the scope, applicability, or configuration of the present disclosure.Rather, the foregoing description of the exemplary embodiments will provide thoseskilled in the art with an enabling description for implementing any of the exemplaryembodiments. Specific details are given in the following description to provide athorough understanding of the embodiments. However, it may be understood by one ofthe ordinary skilled in the art that the embodiments disclosed herein may be practicedwithout these specific details.
[0038] The terminology used herein is for the purpose of describing particularembodiments only and is not intended to be limiting of the disclosure. As used hereinthe description, the singular forms "a", "an", and "the" include plural forms unless thecontext of the invention indicates otherwise.
[0039] The terminology and structure employed herein are for describing, teaching,and illuminating some embodiments and their specific features and elements and donot limit, restrict, or reduce the scope of the present disclosure. Accordingly, unlessotherwise defined, all terms, and especially any technical and / or scientific terms, usedherein may be taken to have the same meaning as commonly understood by one havingordinary skill in the art.
[0040] Further, few of the figures in the accompanying drawings are represented incolor for accurate illustration of results. The color drawings illustrate one or moreresults which cannot be adequately represented in black and white or gray scale. Thecolor drawings are therefore necessary to ensure proper interpretation of the disclosedinvention.
[0041] The present disclosure relates to a system and a method for estimating areliability of a model used for predicting thermal error of a machine tool.
[0042] Embodiments of the present disclosure will be described below in detail withreference to the accompanying drawings. FIG. 1 through FIG. 10, discussed below, andthe one or more embodiments used to describe the principles of the present disclosureare by way of illustration only and should not be construed in any way to limit thescope of the present disclosure. Those skilled in the art will understand that theprinciples of the present disclosure may be implemented in any suitably arrangedsystem or device.
[0043] FIG. 1 illustrates a block diagram of a system 100 for real-time thermal errorprediction and reliability assessment in machine tools, in accordance with anembodiment of the present disclosure.
[0044] As shown in FIG. 1, the system 100 includes a Computer Numerical Control(CNC) controller 102, machine tool 104, an Industrial Internet of Thing (IIoT) module106, and a cloud platform 108 (may also be referred as the "server 108"). The machinetool 104 may include various mechanical and structural components that are subject tothermal effects during machining operations. The present disclosure relates tomonitoring and compensating these thermal effects. The IIoT module 106 may collectreal-time temperature data from one or more hotspot locations on the machine tool 104using integrated temperature sensors. The temperature data reflects thermal conditionof the machine tool during the machining operations and is transmitted to the cloudplatform for further processing.
[0045] The cloud platform 108 includes a processor 108-2, a memory 108-4. Theprocessor 108-2 may include various processing circuitry and communicate with thememory 108-4 via a communication bus 108-6. The processor 108-2 is configured toexecute instructions or a set of instructions stored in the memory 108-4 to performvarious processes. The processor 108-2 may include a general-purpose processor, suchas, for example, and without limitation, a Central Processing Unit (CPU), anApplication Processor (AP), a dedicated processor, a Graphics-only Processing Unitsuch as a Graphics Processing Unit (GPU) or the like, a programmable logic device, orany combination thereof.
[0046] The memory 108-4 stores the set of instructions required by the processor 108-2 of the server 108 for controlling its overall operations. The memory 108-4 mayinclude non-volatile storage elements. Examples of such non-volatile storage elementsmay include magnetic hard discs, optical discs, floppy discs, flash memories, or formsof Electrically Programmable Memories (EPROM) or Electrically Erasable andProgrammable (EEPROM) memories. In addition, the memory 104 may, in someexamples, be considered a non-transitory storage medium. The "non-transitory" storagemedium is not embodied in a carrier wave or a propagated signal. However, the term"non-transitory" should not be interpreted as the memory 108-4 is non-movable. Insome examples, the memory 108-4 may be configured to store larger amounts ofinformation. In certain examples, a non-transitory storage medium may store data thatcan, over time, change (e.g., in Random Access Memory (RAM) or cache). Thememory 108-4 may be an internal st orage unit or an external storage unit of the server108, cloud storage, or any other type of external storage.
[0047] The cloud platform 108 may receive the real-time temperature data. The data isused to train a Thermal Error Prediction Model (TEPM) that estimates thermal-inducedpositional deviations based on the real-time temperature data. The TEPM may refer toa machine learning model that estimates the thermal-induced positional or dimensionaldeviations in precision manufacturing equipment. The cloud platform 108 mayevaluates robustness of the trained TEPM by applying controlled variations to the realtimetemperature data to simulate sensor degradation. The model's outputs using thisdegraded data are compared with measured thermal error values to assess informationloss using a predetermined performance metric. Based on the information loss, thereliability of the TEPM is estimated. When the estimated reliability of the TEPM fallsbelow a predefined threshold, the cloud platform 108 initiates an updating processwherein the TEPM is retrained using continuous machine learning techniques.
[0048] The CNC controller 102 may receive thermal error compensation values fromthe cloud platform 108 and applies these values to the machine tool 104. Feedbacksignals and command signals are exchanged between the CNC controller 102 and themachine tool 104 to adjust axis movement.
[0049] FIG. 2 illustrates a block diagram of the IIoT module 106 of the system 100, inaccordance with an embodiment of the present disclosure. The IIoT module 106comprises multiple interconnected components such as temperature sensor(s) 202, aLiquid-Crystal Display (LCD) 204, a micro-SD 206, an Analog-to-Digital Converter(ADC) (n / 4) 208 (may also be referred as the "ADC 208"), a microcontroller 210, anopto-coupler 212, 4G modem 214, a Wi-Fi module 216, and a power supply 218. Thetemperature sensor(s) 202 are deployed at designated hotspots of the machine tool 104to sense temperature variations in real time. The LCD 204 is a display unit connectedto the microcontroller 210 and configured to visually present thermal data,compensation values, or system status in real time. The microSD 206 may be a memorystorage unit for logging temperature data, and compensation results. The ADC 208 isconfigured to receive analog signals from the temperature sensor(s) 202 and convertthem into digital format. The "n / 4" may indicate that the ADC 208 may accommodatemultiple sensor inputs. In an embodiment, the ADC 208 is used to read the temperatureof up to 16 sensors. The ADC 208 may support up to 8 differential inputs and readswith 16-bit resolution. Further, the ADC 208 is interfaced with core of the IIoT module106 through an I2C interface. Two ADC chips are being used to read temperaturevalues of up to 16 temperature sensors. The read temperature is stored in a processor,where edge processing is done, and processed data is sent to the cloud platform 108using an ESP32-based WIFI interface or a 4G modem (L510). The microcontroller 210is a central processing unit of the IIoT module 106 that receives digitized temperaturedata from the ADC (n / 4) 208, processes the digitized temperature data, and performspredictive thermal error compensation computations using a deployed model. Theopto-coupler 212 may be a component that is used to safely interface themicrocontroller 210 with external control hardware. The 4G Modem 214 and the Wi-Fi module 216 are wireless communication modules configured to transmit data to thecloud platform 108 and may receive updates or model parameters. The 4G modem 214may enables remote connectivity in industrial setups where the Wi-Fi module 216 maynot be reliable. The power supply 218 may provide regulated voltages necessary forthe operation of the microcontroller 210 and connected peripherals, sourced from apower line. In an embodiment, the power supply 218 may provide the regular voltagesof 5V, 3.8V, 3.3V sourced from a 24V power line.
[0050] FIG. 3 illustrates a diagram of a circuit board 300 of the IIoT module 106, inaccordance with an embodiment of the present disclosure. Terminals are provided onthe circuit board 300 to connect Resistance Temperature Detectors (RTDs), thetemperature sensors, at critical hotspots of the machine tool 104 to gather real-timethermal data. Referring to Fig. 3, the circuit board 300 includes embedded Wi-Fifunctionality for wireless communication with the cloud platform 108 to support realtimemodel execution and data upload. In an embodiment, core of the circuit board 300is an arm cortex M7 microcontroller with 16kB I-cache and 16kB data cache. Thecircuit board 300 has a Floating-Point Unit (FPU), high-speed USB, and otherperipherals. The FPU may help in dealing with floating point arithmetic. In animplementation, the circuit board may have optional support for 3.2-inch RGB LEDwith 320x480 pixel resolution. In a designated portion of the circuit board 300, themicrocontroller and memory components are housed to execute deployed TEPM andfor managing input processing logic.
[0051] The circuit board 300 comprises an output terminal from where compensationvalues computed by the microcontroller 210 are transmitted to the CNC controller 102.
[0052] FIG. 4A illustrates a schematic diagram of thermal sensor locations on themachine tool 104, in accordance with an embodiment of the present disclosure. FIG.4B illustrates an experimental setup implemented for measuring thermal drift at a ToolCentre Point (TCP) caused by temperature variations during the machine operation, inaccordance with an embodiment of the present disclosure.
[0053] Referring to FIG. 4A, a slant bed CNC turning center is selected for evaluationof the reliability of the TEPMs. In an embodiment, six hotspot locations on the machinetool 104 may be identified using infrared imaging. The RTDs are mounted at the sixidentified hot spots (T1 to T6). In an embodiment, the identified hotspots are ambienttemperature T1, reference bed T2, spindle rear T3, coolant tank wall T4, transformerbed T5, and spindle front T6. Referring to FIG. 4B, a cylindrical target made of Invarmaterial is mounted in a chuck, and five current sensors are mounted on the machinetool post in an Invar-made fixture to measure a tool center point drift during air-cuttingtrials. The current sensors are positioned in X, Y, and Z directions to measure thedeviation or drift of the TCP over time. These readings are used in training the TEPMand evaluating the performance of the TEPM during a degradation simulation. In anembodiment, the invar material is selected due to its negligible thermal expansioncoefficient, thereby ensuring accurate and stable measurements unaffected by thermaldeformation of the fixture itself.
[0054] FIG. 5 illustrates the method for estimating the reliability of the TEPM, inaccordance with an embodiment of the present disclosure.
[0055] At step 502, the processor 108-2 may receive the temperature data acquiredfrom temperature sensors 202 mounted at the critical hotspot locations on the machinetool 104. These temperature readings are collected by the IIoT module 106 andtransmitted in real time to the processor 108-2.
[0056] At step 504, the processor 108-2 may measure the thermal error value at theTCP. In an embodiment, the measured thermal error value is obtained throughexperimental measurements using displacement sensors such as eddy current sensors,which detect structural deformation or drift of the TCP caused by thermal effects.
[0057] At step 506, the processor 108-2 may train the TEPM by correlating thereceived temperature data with the experimentally measured thermal error at the TCP.The TEPM model is periodically retrained or updated to adapt to sensor degradation orprocess drift.
[0058] At step 508, the processor 108- 2 may generate degraded temperature data byapplying predefined variations to the received temperature data. The predefinedvariations include simulation of a noise or a predefined failure criterion in the receivedtemperature data.
[0059] At step 510, the processor 108-2 may input the degraded temperature data intothe trained TEPM. The processor 108-2 uses this data to generate predicted thermalerror values.
[0060] At step 512, the processor 108-2 may determine performance metrics databased on a difference between the predicted thermal error value and the measuredthermal error value. One or more performance metrics is computed such as MeanAbsolute Error (MAE), Root Mean Square Error (RMSE), or Error Loss Percentage.
[0061] At step 514, the processor 108-2 may estimate the reliability of the TEPM underdegraded sensor conditions using the computed performance metrics data. In anembodiment, if the value associated with performance metrics data drops below thepredefined threshold, this indicates information loss, prompting the system to initiateTEPM retraining or adjustment. The TEPM is updated based on the information lossand upon the updating of the TEPM, the reliability of the model is re-estimated untilthe estimated reliability meets or exceeds a predefined reliability threshold.
[0062] FIG. 6 illustrates a detailed flowchart depicting the method for estimating thereliability of the TEPM and the upgradation of the TEPM using continuous machinelearning framework, in accordance with an embodiment of the present disclosure.
[0063] At step 602, the method begins with an experimentation phase. Theexperimentation is done to collect the temperature data and corresponding thermal errorvalues at the TCP. In an embodiment, Infrared (IR) thermal imaging is used to performthermal sensitivity analysis of the machine tool 104. Then several thermal hotspots (n)are identified at structural locations on the machine tool 104. The RTDs are thenmounted at these identified hotspots. In one or more embodiments, to simulate thethermal effects during the machining operation, air-cutting trials are performed. Duringthe air-cutting trials, eddy current sensors are employed at machine tool post to measurethe TCP drift, which indicates thermal deformation. The temperature data, and themeasured thermal error values are the datasets. The datasets are processed and passedto the TEPM. Most of the dataset is used for training and validation of the TEPM andremaining dataset is used for TEPM evaluation and testing.
[0064] At step 604, once the TEPM is trained, the reliability of the TEPM is evaluatedunder degraded sensor conditions. In an embodiment, predefined noise levels areapplied to the temperature data to simulate noisy sensors. In another embodiment,sensor values are replaced with nulls or erroneous values based on predefined sensorfailure criteria. This creates a set of degraded input temperature data. For each degradedsensor scenario, the trained TEPM is used to predict the thermal error values. TheInformation loss is computed as a metric quantifying the difference between themeasured thermal error and the predicted thermal error. This may include the RMSE,and Kullback-Leibler Divergence (KL Divergence). A threshold RMSE value is predefinedto signify acceptable degradation. If a predicted RMSE under sensordegradation is below the predefined threshold, it indicates model failure. In one or moreembodiments, methods like, but not limited to, First Order Reliability Method(FORM), Monte Carlo Simulation (MCS), or Importance Sampling Method (ISM) areused in the reliability estimation to compute the probability of the failure and theestimated reliability value of the TEPM.
[0065] At step 606, if the estimated reliability of the model falls below the predefinedthreshold, the TEPM undergoes the updating process. Updated sensor data from themachine tool is acquired, including an updated temperature data and an updated TCPthermal error value. The system monitors information loss metrics. When loss valuesindicate degradation, the TEPM is retrained using the updated values. Upon theupdating of the model, the reliability of the TEPM is re-estimated until the estimatedreliability meets or exceeds the predefined reliability threshold.
[0066] FIG. 7A illustrates a variation in sensor data under different degradation states,in accordance with an embodiment of the present disclosure. FIG. 7B illustrates acomparison of the predicted thermal error using different machine learning models, inaccordance with an embodiment of the present disclosure.
[0067] FIG. 7A illustrates the variation in temperature readings captured from anexemplary sensor T3 under various simulated degradation states. The actualtemperature profile ("Actual T3 Data") is compared against temperature profilesmodified with additive noise models of different intensities such as Low Noise (LN),Medium Noise (MN), and High Noise (HN), and a fully failed state. X-axis representsa number of data points, and y-axis represents temperature in degrees Celsius. FIG. 7Aillustrates distortion in temperature data upon the sensor degradation, with increasednoise levels causing greater divergence from the actual sensor data trajectory.
[0068] FIG. 7B presents a comparison between the actual thermal error and thepredicted thermal errors obtained using exemplary four machine learning models. Inone or more embodiments, various machine learning models may be used such as, butnot limited to, Multiple Linear Regression (MLR), Support Vector Regression (SVR),Feed-Forward Neural Network (FFNN), and Long Short-Term Memory (LSTM). Thex-axis corresponds to the number of data points, while the y-axis represents thermalerror in micrometers (μm). FIG. 7B depicts that the LSTM model provides closerapproximation to the actual thermal error across all time steps, while traditional modelssuch as MLR show higher deviation, particularly around data transition points.
[0069] FIG. 8A illustrates the reliability of prediction models under the sensordegradation, in accordance with an embodiment of the present disclosure. FIG. 8Billustrates probability distribution of the prediction error under the sensor degradation,in accordance with an embodiment of the present disclosure.
[0070] FIG. 8A depicts bar plots representing the estimated reliability of four TEPMssuch as the MLR, the SVR, the FFNN, and the LSTM under varying numbers ofdegraded temperature sensors (1, 2, and 3). The reliability estimation is carried outusing three probabilistic methods such as the FORM, the MCS, and the ISM. The yaxisdenotes model reliability in percentage. FIG. 8B displays Probability DensityFunctions (PDFs) of the RMSE for different prediction models under varying levels ofsensor degradation. Each subfigure presents a distribution of the RMSE values forMLR, SVR, FFNN, and LSTM, and a vertical dashed line indicates a threshold RMSEvalue (8 μm) beyond which model performance is considered unacceptable.
[0071] FIG. 9A illustrates prediction error analysis across the sensor degradationstates, in accordance with an embodiment of the present disclosure. FIG. 9B illustratesan exemplary KL divergence analysis across the sensor degradation states, inaccordance with an embodiment of the present disclosure.
[0072] FIG. 9A illustrates the thermal prediction error (in μm) for the sensors T1-T6under various degradation states, including the LN, the MN, the HN, and the completesensor failure. A degradation impact is plotted for each temperature sensor. The graphsdepict a trend of increasing prediction error with sensor degradation severity. FIG. 9Bdepicts the KL divergence between the probability distribution of the predicted thermalerror and the distribution obtained under actual conditions. The divergence is measuredfor each sensor under increasing degradation states (LN, MN, HN, Failed). The x-axisrepresents the sensor states, and the y-axis represents a KL divergence metric. HigherKL divergence signifies greater deviation from the true distribution, thereby indicatingmodel uncertainty and reduced performance.
[0073] FIG. 10 illustrates an effectiveness of model upgradation in degraded sensorenvironment, in accordance with an embodiment of the present disclosure. FIG.10demonstrates the performance evolution of multiple machine learning models beforeand after the upgradation in response to the degraded sensor data. A top subplot showsmodel Mean Squared Error (MSE) on new degraded data before the upgradation. Themiddle subplot shows the MSE on the same data after the model upgradation. Thebottom subplot measures a percentage improvement in prediction accuracy followingthe upgradation.
[0074] Now, referring to the technical abilities and advantageous effect of the presentdisclosure, the one or more embodiments provides various operational advantagesdescribed below. The present disclosure describes the TEPM incorporating advancedmachine learning models which demonstrates the reliability even when one or moretemperature sensors degrade or fail. The present disclosure uses reliabilityquantification methods such as the FORM, the MCS, and the ISM to assess andcompare failure probabilities of different models. This enables risk-aware modelselection and justifies deployment of more robust TEPM architectures underuncertainty. Unlike conventional accuracy metrics, the present disclosure offersprobabilistic distribution analysis of the RMSE across varying degrees of the sensordegradation. The TEPM is designed to work under different sensor configurations,from healthy states to partial or complete sensor failures. Reliability insights derivedfrom the TEPM can trigger autonomous model upgradation or re-learning processes,forming a basis for intelligent maintenance systems.
[0075] Those skilled in the art will appreciate that the methodology described hereinin the present disclosure may be carried out in other specific ways than those set forthherein in the above disclosed embodiments without departing from essentialcharacteristics and features of the present invention. The above-describedembodiments are therefore to be construed in all aspects as illustrative and notrestrictive.
[0076] The drawings and the forgoing description give examples of embodiments.Those skilled in the art will appreciate that one or more of the described elements maywell be combined into a single functional element. Alternatively, certain elements maybe split into multiple functional elements. Elements from one embodiment may beadded to another embodiment. For example, orders of processes described herein maybe changed and are not limited to the manner described herein. Any combination of theabove features and functionalities may be used in accordance with one or moreembodiments.
[0077] In the present disclosure, each of the embodiments has been described withreference to numerous specific details which may vary from embodiment toembodiment. The foregoing description of the specific embodiments disclosed hereinmay reveal the general nature of the embodiments herein that others may, by applyingcurrent knowledge, readily modify and / or adapt for various applications such specificembodiments without departing from the generic concept, and, therefore, suchadaptations and modifications are intended to be comprehended within the meaning ofthe disclosed embodiments. It is to be understood that the phraseology or terminologyemployed herein is for the purpose of description and is not limited in scope.
Claims
1. A method (500) for estimating a reliability of a model used for predicting thermal error of a machine tool (104), the method comprising: receiving (502), from one or more temperature sensors, temperature data associated with one or more positions on the machine tool; measuring (504), based on the received temperature data, a thermal error value at a tool center point of the machine tool (104); training (506) the model using the received temperature data and the measured thermal error value; generating (508) degraded temperature data by applying predefined variations to the received temperature data to indicate deterioration in the one or more temperature sensors; providing (510) the degraded temperature data to the trained model to obtain a predicted thermal error value; determining (512) performance metrics data based on a difference between the predicted thermal error value and the measured thermal error value; and estimating (514) the reliability of the model based on the determined performance metrics data.
2. The method (500) as claimed in claim 1, further comprising: determining whether a value associated with the performance metrics data falls below a threshold value; determining an information loss based on the determination that the value associated with the performance metrics data falls below the threshold value; and updating the model based on the determination of the information loss.
3. The method (500) as claimed in claim 2, further comprising: re-estimating, upon the updating of the model, the reliability of the model until the estimated reliability meets or exceeds a predefined reliability threshold.
4. The method (500) as claimed in claim 1, wherein the predefined variations include simulation of a noise or a predefined failure criterion in the received temperature data.
5. The method (500) as claimed in claim 1, wherein the performance metric data comprises values of functions selected from one or more of Root Mean Square Error (RMSE) and Mean Absolute Error (MEA).
6. A system (100) for estimating a reliability of a model used for predicting thermal error of a machine tool (104), the system (100) comprising a server (108) including at least one processor (108-2), wherein the at least one processor (108-2) is configured to: receive, from one or more temperature sensors, temperature data associated with one or more positions on the machine tool (104); measure, based on the received temperature data, a thermal error value at a tool center point of the machine tool (104); train the model using the received temperature data and the measured thermal error value; generate degraded temperature data by applying predefined variations to the received temperature data to indicate deterioration in the one or more temperature sensors; provide the degraded temperature data to the trained model to obtain a predicted thermal error value; determine performance metrics data based on a difference between the predicted thermal error value and the measured thermal error value; and estimate the reliability of the model based on the determined performance metrics data.
7. The system (100) as claimed in claim 6, wherein the at least one processor (108- 2) is further configured to: determine whether a value associated with the performance metrics data falls below a threshold value; determine an information loss based on the determination that the value associated with the performance metrics data falls below the threshold value; and update the model based on the determination of the information loss.
8. The system (100) as claimed in claim 7, wherein the at least one processor (108- 2) is further configured to re-estimate, upon the updating of the model, the reliability of the model until the estimated reliability meets or exceeds a predefined reliability threshold.
9. The system (100) as claimed in claim 6, wherein the predefined variations include simulation of a noise or a predefined failure criterion in the received temperature data.
10. The system (100) as claimed in claim 6, wherein the performance metric data comprises values of functions selected from one or more of Root Mean Square Error (RMSE) and Mean Absolute Error (MEA).
