Advanced thermal compensation of mechanical processes

Patent metadata

Field	Value
Publication number	US-11550291-B2
Application number	US-202017112836-A
Country	US
Kind code	B2
Filing date	Dec 4, 2020
Priority date	Dec 20, 2019
Publication date	Jan 10, 2023
Grant date	Jan 10, 2023

Abstract

A computer program product and to a method for compensating thermal errors in a mechanical process, the mechanical process in particular provided by a mechanical device such as a coordinate measuring machine, a tooling machine or an articulated robot arm. Thermal errors arise due to thermal disturbances affecting the mechanical process, wherein thermal disturbances may arise from environmental influences affecting the mechanical process or from internally generated changing temperature distributions.

First claim

What is claimed is:

1. A computer program product for compensating errors of a mechanical process due to thermal disturbances, the computer program product having a non-transitory machine-readable storage medium that includes computer-executable instructions that when executed by a computer is configured to receive as input:
a set of sensor input data points from a sensor input data point space, each sensor input data point comprising data from a set of sensors, wherein each sensor provides a sensor sequence of sensor values at sensor sequence times, and each sensor input data point corresponds to an evolution of the mechanical process, wherein different sensor input data points correspond to different evolutions of the mechanical process;
a current sensor input data point;
a set of thermal error compensation states of the mechanical process, wherein each thermal error compensation state comprises a thermal error compensation state sequence of thermal error compensation vectors, wherein the thermal error compensation vectors are provided at thermal error compensation state sequence times;
a model of the mechanical process, the model comprising a set of free parameters and a set of internal parameters, the assignment of free parameters and internal parameters determining temporal and spatial evolution of the model, the free parameters having pre-assigned values, wherein the free parameters are parameters that are determined through a calibration step;
constraints on the temporal and spatial evolution of the model in the form of possible sets of values which the free parameters can take and in the form of sets of relationships between the internal parameters of the model;
an abstract representation embodied as a function mapping an input space onto the space of thermal error compensation states, wherein the input space comprises at least the sensor input data point space, the function having a set of function parameters; and
a representation input function with assigned function parameter values, the representation input function corresponding in form to the abstract representation,

wherein the computer program product possesses two modes of operation, wherein in a training mode a training data set comprising empirical training data or synthetic training data is used, wherein the empirical training data are comprised of the set of sensor input data points as empirical training data input and the set of thermal error compensation states as empirical training data output, and wherein the synthetic training data are constructed by way of simulations carried out using the model of the mechanical process and the constraints, the synthetic training data comprising a synthetic simulated set of sensor input data points as synthetic training data input and a synthetic simulated set of thermal error compensation states as synthetic training data output,
wherein the training data set comprising the empirical or synthetic training data is used for adapting the representation input function by adapting the function parameters, an inference mode being configured to use the adapted representation input function, wherein the adapted representation input function is invoked with at least the current sensor input data point to calculate a current thermal error compensation state used for compensating the thermal errors arising during the execution of the mechanical process.

2. The computer program product according to claim 1 , wherein at least one sensor input data point of the set of sensor input data points comprises initial state information about the corresponding mechanical process at an initial state time, and/or at least one sensor input data point of the set of sensor input data points comprises nominal information used for controlling the mechanical process, or the current sensor input data point comprises current thermal error compensation state sequence times, wherein the current thermal error compensation state provided by the adapted representation input function is provided at the current thermal error compensation state sequence times.

3. The computer program product according to claim 1 , wherein the sensor sequences of sensor values used as input to the abstract representation are aligned in time, wherein corresponding sensor sequence values of a first and a second sensor sequence are obtained at the same time during the evolution of the mechanical process, the set of times at which sensor sequence values are provided being regularly sampled or event-based and/or user-triggered, and the thermal error compensation state sequence times are aligned with the sensor sequence times.

4. The computer program product according to claim 1 , wherein the number of function parameters of the abstract representation is smaller than the number of free parameters and internal parameters of the model of the mechanical process.

5. The computer program product according to claim 1 , wherein the set of thermal error compensation states provided as input to the computer program product provides information about thermal error compensation only, and the synthetic simulated set of thermal error compensation states is configured to provide information about thermal error compensation only, wherein compensations for other disturbances affecting the mechanical process are substantially separable from thermal error compensation.

6. The computer program product according to claim 1 , wherein the constraints provided to the computer program product reflect specific requirements and boundary conditions, the constraints limiting the space of possible evolutions of the mechanical process, wherein the constraints are reflected in the synthetic training data, the synthetic training data generated according to the constraints, the model of the mechanical process and a set of data-generated constraints, wherein the data-generated constraints are derived from the set of sensor input data points.

7. The computer program product according to claim 1 , wherein the number of generated synthetic training data is larger than the number of function parameters of the abstract representation, wherein the synthetic training data comprise examples for evolutions of the mechanical process compatible with the constraints, the number of synthetic training data per nominal evolution compatible with a constraint being related to the expected frequency of occurrence of said nominal evolution of the mechanical process, and the training data set is only comprised of synthetic training data.

8. The computer program product according to claim 1 , wherein a fixed number of sensor sequence elements is considered by the abstract representation, wherein the fixed number is the same for all sensors providing information to the computer program product, wherein the abstract representation is configured to weight the sensor sequence elements in case of temporal irregularity, temporal irregularity understood with respect to the sensor sequence times at which the sensor sequence elements were obtained, the weighting pattern taking the irregularity of the input to the abstract representation into account.

9. The computer program product according to claim 1 , wherein the abstract representation is embodied as a recurrent neural network, or as a convolutional neural network, or as a feedforward neural network, or as ridge regression, or as a feedforward neural network, or as polynomial of fixed degree or as an algebraic expression or elementary function, or the model of the mechanical process is given in the form of a finite element method, or by a combination of a numerical model with symbolic equations.

10. The method for providing indication about the physical sources of thermal errors

Assignees

• Hexagon Technology Ct Gmbh

Inventors

• Iseli Claudio
• Reimann Bernd
• Schlenkrich-Erasmus Michael
• Schneider Jürgen
• Mentl Katrin
• Burgstaller Roland
• Lamping Frank
• Heili Alexandre
• Rana Asif
• Aebischer Beat

Classifications

• G01B21/045

  Correction of measurements (G01B9/02055 takes precedence) · CPC title

• G05B9/02

  electric · CPC title

• G05B19/404 Primary

  characterised by control arrangements for compensation, e.g. for backlash, overshoot, tool offset, tool wear, temperature, machine construction errors, load, inertia (G05B19/19, G05B19/41 take precedence) · CPC title

• G05B2219/35015

  Calculate production compensation, heat shrinkage, overetching · CPC title

• G01B5/0014 Primary

  due to temperature (on machine tools B23Q11/0003) · CPC title