Gm Global Technolgoy Operations, Inc.

University of Michigan

Michael Anthony Wincek

Jeffrey A Abell

John Patrick Spicer

Debejyo Chakraborty

Hui Wang

Wayne Cai

Jennifer Bracey

Paul F Spacher

Shixin Jack Hu

Daniel C Hutchinson

Peter W Tavora

Paul J Boor

Jason C H Tang

Jeffrey S Davis

Ronald L Reardon

Shawn Utz

Diana Marie Wegner

An automated method for discovering features in a repeatable process includes measuring raw time series data during the process using sensors. The time series data describes multiple parameters of the process. The method includes receiving, via a first controller, the time series data from the sensors, and stochastically generating candidate features from the raw time series data using a logic block or blocks of the first controller. The candidate features are predictive of a quality of a work piece manufactured via the repeatable process. The method also includes determining, via a genetic or evolutionary programming module, which generated candidate features are most predictive of the quality of the work piece, and executing a control action with respect to the repeatable process via a second controller using the most predictive candidate features. A system includes the controllers, the programming module, and the sensors.

Automated stochastic method for feature discovery and use of the same in a repeatable process

A method includes receiving, during a vibration welding process, a set of sensory signals from a collection of sensors positioned with respect to a work piece during formation of a weld on or within the work piece. The method also includes receiving control signals from a welding controller during the process, with the control signals causing the welding horn to vibrate at a calibrated frequency, and processing the received sensory and control signals using a host machine. Additionally, the method includes displaying a predicted weld quality status on a surface of the work piece using a status projector. The method may include identifying and display a quality status of a suspect weld. The laser projector may project a laser beam directly onto or immediately adjacent to the suspect welds, e.g., as a red, green, blue laser or a gas laser having a switched color filter.

Quality status display for a vibration welding process

A system includes a host machine and a status projector. The host machine is in electrical communication with a collection of sensors and with a welding controller that generates control signals for controlling the welding horn. The host machine is configured to execute a method to thereby process the sensory and control signals, as well as predict a quality status of a weld that is formed using the welding horn, including identifying any suspect welds. The host machine then activates the status projector to illuminate the suspect welds. This may occur directly on the welds using a laser projector, or on a surface of the work piece in proximity to the welds. The system and method may be used in the ultrasonic welding of battery tabs of a multi-cell battery pack in a particular embodiment. The welding horn and welding controller may also be part of the system.

A method for monitoring and controlling a vibration welding system includes collecting sensory data during formation of a welded joint using sensors positioned with respect to welding interfaces of a work piece. A host machine extracts a feature set from a welding signature collectively defined by the sensory data, compares and correlates the feature set with validated information in a library, and executes a control action(s) when the present feature set insufficiently matches the information. A welding system includes a sonotrode, sensors, and the host machine. The host machine is configured to execute the method noted above.

Method and system for online quality monitoring and control of a vibration welding process

A system includes host and learning machines. Each machine has a processor in electrical communication with at least one sensor. Instructions for predicting a binary quality status of an item of interest during a repeatable process are recorded in memory. The binary quality status includes passing and failing binary classes. The learning machine receives signals from the at least one sensor and identifies candidate features. Features are extracted from the candidate features, each more predictive of the binary quality status. The extracted features are mapped to a dimensional space having a number of dimensions proportional to the number of extracted features. The dimensional space includes most of the passing class and excludes at least 90 percent of the failing class. Received signals are compared to the boundaries of the recorded dimensional space to predict, in real time, the binary quality status of a subsequent item of interest.

Binary classification of items of interest in a repeatable process

A vibration welding system includes vibration welding equipment having a welding horn and anvil, a host device, a check station, and a robot. The robot moves the horn and anvil via an arm to the check station. Sensors, e.g., temperature sensors, are positioned with respect to the welding equipment. Additional sensors are positioned with respect to the check station, including a pressure-sensitive array. The host device, which monitors a condition of the welding equipment, measures signals via the sensors positioned with respect to the welding equipment when the horn is actively forming a weld. The robot moves the horn and anvil to the check station, activates the check station sensors at the check station, and determines a condition of the welding equipment by processing the received signals. Acoustic, force, temperature, displacement, amplitude, and/or attitude/gyroscopic sensors may be used.

Automatic monitoring of vibration welding equipment

A system includes host and learning machines in electrical communication with sensors positioned with respect to an item of interest, e.g., a weld, and memory. The host executes instructions from memory to predict a binary quality status of the item. The learning machine receives signals from the sensor(s), identifies candidate features, and extracts features from the candidates that are more predictive of the binary quality status relative to other candidate features. The learning machine maps the extracted features to a dimensional space that includes most of the items from a passing binary class and excludes all or most of the items from a failing binary class. The host also compares the received signals for a subsequent item of interest to the dimensional space to thereby predict, in real time, the binary quality status of the subsequent item of interest.

Growing community of inventors

Rochester, MI, United States of America

Michael Anthony Wincek