Monitoring machine health and product quality enables predictive maintenance that optimizes repairs to minimize factory downtime. Data-driven intelligent manufacturing often relies on probabilistic techniques with intractable distributions. For example, generative models of data distributions can balance fault classes with synthetic data, and sampling the posterior distribution of hidden model parameters enables prognosis of degradation trends. Normalizing flows can address these problems while avoiding the training instability or long inference times of other generative Deep Learning (DL) models like Generative Adversarial Networks (GAN), Variational Autoencoders (VAE), and diffusion networks. To evaluate normalizing flows for manufacturing, experiments are conducted to synthesize surface defect images from an imbalanced data set and estimate parameters of a tool wear degradation model from limited observations. Results show that normalizing flows are an effective, multi-purpose DL architecture for solving these problems in manufacturing. Future work should explore normalizing flows for more complex degradation models and develop a framework for likelihood-based anomaly detection. Code is available at https://github.com/uky-aism/flows-for-manufacturing.
Smart manufacturing systems are considered the next generation of manufacturing applications. One important goal of the smart manufacturing system is to rapidly detect and anticipate failures to reduce maintenance cost and minimize machine downtime. This often boils down to detecting anomalies within the sensor data acquired from the system which has different characteristics with respect to the operating point of the environment or machines, such as, the RPM of the motor. In this paper, we analyze four datasets from sensors deployed in manufacturing testbeds. We detect the level of defect for each sensor data leveraging deep learning techniques. We also evaluate the performance of several traditional and ML-based forecasting models for predicting the time series of sensor data. We show that careful selection of training data by aggregating multiple predictive RPM values is beneficial. Then, considering the sparse data from one kind of sensor, we perform transfer learning from a high data rate sensor to perform defect type classification. We release our manufacturing database corpus (4 datasets) and codes for anomaly detection and defect type classification for the community to build on it. Taken together, we show that predictive failure classification can be achieved, paving the way for predictive maintenance.
more » « less- Award ID(s):
- 2016704
- NSF-PAR ID:
- 10495797
- Publisher / Repository:
- MDPI
- Date Published:
- Journal Name:
- Sensors
- Volume:
- 23
- Issue:
- 1
- ISSN:
- 1424-8220
- Page Range / eLocation ID:
- 486
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
Abstract -
In many modern enterprises, factory managers monitor their machinery and processes to prevent faults and product defects, and maximize the productivity and efficiency. Asset condition, product quality and system productivity monitoring consume some 40-70% of the production costs. Oftentimes, resource constraints have prevented the adoption and implementation of these practices in small businesses. Recent evolution of manufacturing-as-a-service and increased digitalization opens opportunities for small and medium scale companies to adopt smart manufacturing practices, and thereby surmount these constraints. Specifically, sensor wrappers that delineate the specifications of sensor integration into manufacturing machinery, with appropriate edge-cloud computing and communication architecture can provide even small businesses with a real-time data pipeline to monitor their manufacturing machines. However, the data in itself is difficult to interpret locally. Additionally, proprietary standards and products of the various components of a sensor wrapper make it difficult to implement a sensor wrapper schema. In this paper, we report an open-source method to integrate sensors into legacy manufacturing equipment and hardware. We had implemented this pipeline with off-the-shelf sensors to a polisher (from Buehler), a shaft grinding machine (from Micromatic), and a hybrid manufacturing machine (from Optomec), and used hardware and software components such as a National Instruments Data Acquisition (NI-DAQ) module to collect and stream live data. We evaluate the performance of the data pipeline as it connects to the Smart Manufacturing Innovation Platform (SMIP)—web-based data ingestion platform part of the Clean Energy Smart Manufacturing Innovation Institute (CESMII), a U.S. Department of Energy-sponsored initiative—in terms of data volume versus latency tradeoffs. We demonstrate a viable implementation of Smart Manufacturing by creating a vendor-agnostic web dashboard that fuses multiple sensors to perform real-time performance analysis with lossless data integrity.more » « less
-
The application of cutting-edge technologies such as AI, smart sensors, and IoT in factories is revolutionizing the manufacturing industry. This emerging trend, so called smart manufacturing, is a collection of various technologies that support decision-making in real-time in the presence of changing conditions in manufacturing activities; this may advance manufacturing competitiveness and sustainability. As a factory becomes highly automated, physical asset management comes to be a critical part of an operational life-cycle. Maintenance is one area where the collection of technologies may be applied to enhance operational reliability using a machine condition monitoring system. Data-driven models have been extensively applied to machine condition data to build a fault detection system. Most existing studies on fault detection were developed under a fixed set of operating conditions and tested with data obtained from that set of conditions. Therefore, variability in a model’s performance from data obtained from different operating settings is not well reported. There have been limited studies considering changing operational conditions in a data-driven model. For practical applications, a model must identify a targeted fault under variable operational conditions. With this in mind, the goal of this paper is to study invariance of model to changing speed via a deep learning method, which can detect a mechanical imbalance, i.e., targeted fault, under varying speed settings. To study the speed invariance, experimental data obtained from a motor test-bed are processed, and time-series data and time–frequency data are applied to long short-term memory and convolutional neural network, respectively, to evaluate their performance.more » « less
-
Digital Fringe Projection for Interlayer Print Defect Characterization in Directed Energy Depositionmore » « less
Abstract Directed Energy Deposition (DED) is one of the main additive manufacturing (AM) families, enabling the fabrication of multi-material parts with high material addition rates. However, the incremental nature of DED fabrication makes it prone to local defect formation due to process condition fluctuations. Known for its rapid and precise 3D surface measurement capabilities, digital fringe projection (DFP) was previously demonstrated in process monitoring for powder bed AM. This study brings DFP to the DED process through development of a custom motor stage system and validates its effectiveness in assessing surface topography and build height measurement. Measurements were taken on both correctly deposited builds and builds with off-nominal deposition conditions, where the system was able to detect pitting as small as 0.425 mm in the lateral size and 0.154 mm in depth in the case of reduced laser energy. This work paves the way for future machine learning-enabled interlayer defect identification, classification, and healing via altering subsequent processing settings.
-
Machine learning algorithms are increasingly used for inference and decision-making in embedded systems. Data from sensors are used to train machine learning models for various smart functions of embedded and cyber-physical systems ranging from applications in healthcare, autonomous vehicles, and national security. However, recent studies have shown that machine learning models can be fooled by adding adversarial noise to their inputs. The perturbed inputs are called adversarial examples. Furthermore, adversarial examples designed to fool one machine learning system are also often effective against another system. This property of adversarial examples is calledmore » « less
adversarial transferability and has not been explored in wearable systems to date. In this work, we take the first stride in studying adversarial transferability in wearable sensor systems from four viewpoints: (1) transferability between machine learning models; (2) transferability across users/subjects of the embedded system; (3) transferability across sensor body locations; and (4) transferability across datasets used for model training. We present a set of carefully designed experiments to investigate these transferability scenarios. We also propose a threat model describing the interactions of an adversary with the source and target sensor systems in different transferability settings. In most cases, we found high untargeted transferability, whereas targeted transferability success scores varied from 0% to 80%. The transferability of adversarial examples depends on many factors such as the inclusion of data from all subjects, sensor body position, number of samples in the dataset, type of learning algorithm, and the distribution of source and target system dataset. The transferability of adversarial examples decreased sharply when the data distribution of the source and target system became more distinct. We also provide guidelines and suggestions for the community for designing robust sensor systems.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.3390/s23010486
