Machine Learning Engineer, Predictive Maintenance

AssetWatch serves global manufacturers by powering manufacturing uptime through condition monitoring and predictive maintenance solutions. The Machine Learning Engineer will develop and deploy end-to-end machine learning solutions for predictive maintenance, using sensor and machine-context data to diagnose faults, detect anomalies, and assess machinery health. The role also architects scalable AWS-based ML systems, builds distributed data pipelines, and collaborates with MLOps, Data Engineering, and domain experts.

Responsibilities:

• Extract, preprocess, and analyze data from various sensor sources, primarily machine vibration, while also working with temperature, electrical, image, and machine-context data to identify and enhance diagnostic features and fault-specific signatures
• Determine the appropriate modeling techniques for each problem, including classical Machine Learning, supervised and unsupervised learning, anomaly detection, Artificial Neural Networks, Deep Learning, signal-processing methods, and hybrid or rules-based approaches
• Train, validate, and deploy predictive maintenance models to accurately identify and predict machinery faults such as bearing, gearbox, shaft, motor, pump, fan, and belt faults. This may include CNNs, RNNs, LSTMs, attention-based models, and other time-series or representation-learning approaches
• Develop and evaluate time-domain, frequency-domain, and contextual features, including harmonics, sidebands, envelope-spectrum features, running-speed characteristics, bearing and gear-mesh fault frequencies, and other fault-specific evidence
• Act as the solution architect for assigned ML initiatives: define the end-to-end design — data sources and contracts, batch vs. near-real-time execution, feature computation strategy, storage layout (S3 partitioning, Timestream, Aurora/MySQL), orchestration, inference pattern, and cost/scalability tradeoffs — and document the design before large-scale execution, since backfills across tens of thousands of assets are expensive to rerun
• Build and scale distributed data and feature pipelines over large historical sensor datasets using Spark (EMR, Glue, or SageMaker Processing), Athena, and Parquet-based data layouts, with attention to partitioning, throughput, and compute cost
• Design and run data mining and annotation workflows that connect Condition Monitoring Engineers with Data Science — including case discovery, labeling interfaces and conventions, fault-type and time-window labeling, label QA, and turning expert feedback into reusable supervised training data
• Build and maintain benchmark, training, validation, and regression datasets using real machine cases; work with Condition Monitoring Engineers and other domain experts to establish reliable ground truth and measurable model acceptance criteria
• Design experiments and compare new algorithms against existing production methods, investigating false positives, false negatives, suppression behavior, data-quality issues, and performance across different machines, operating conditions, sampling configurations, and sensor characteristics
• Develop robust, modular, reusable, and maintainable ML software in Python, including production-quality model code, shared libraries, configuration, unit and regression tests, code reviews, refactoring, debugging, and version-controlled development workflows
• Use modern AI-assisted and agentic coding approaches to accelerate data preparation, feature engineering, model prototyping, testing, debugging, pipeline development, and documentation; develop or integrate LLM- and agent-based decision-support capabilities where they provide measurable value
• Collaborate with the research and engineering team to constantly refine and improve model architectures, algorithms, and signal-processing approaches, ensuring high accuracy, explainability, scalability, and maintainability
• Work closely with MLOps and Data Engineering teams to ensure smooth deployment of models, feature pipelines, data contracts, and other ML solutions in production environments, including CI/CD for ML, model versioning and registry, IaC (CloudFormation/CDK/Terraform), containerization (Docker/ECR), and production monitoring and alerting via CloudWatch, and participate in production validation and troubleshooting
• Stay updated with the latest advancements in Machine Learning, Deep Learning, Signal Processing, agentic AI, and industrial condition monitoring, ensuring our solutions remain at the forefront of the industry
• Present findings, model performance, strategies, tradeoffs, and solutions to other teams and stakeholders in a clear and concise manner, ensuring that insights drive actionable outcomes

Requirements:

• Master's or Ph.D. in Mechanical Engineering, Electrical Engineering, Computer Science/Engineering, or a related field
• Proven experience in Predictive Maintenance and Condition Monitoring with a focus on Signal Processing techniques such as FFT, Short Time Fourier Transform, Time Synchronous Averaging, Wavelet Transform, Spectral Kurtosis, Spectral Correlation, Envelope Analysis, Hilbert Transform, and related methods
• Strong foundations in Machine Learning and model development, including classical Machine Learning algorithms, supervised and unsupervised learning, anomaly detection, feature engineering, Artificial Neural Networks, Convolutional Neural Networks, Sequence-based Models such as RNNs and LSTMs, and Attention Mechanisms
• Strong understanding of time-series modeling and evaluation, including the ability to select appropriate techniques based on the physical problem, data characteristics, available ground truth, and operational requirements
• Hands-on experience architecting and delivering production ML solutions on AWS is required, including several of: SageMaker (Training, Processing, Pipelines, Model Registry, Endpoints), S3, Athena, AWS Glue, EMR, Lambda, Step Functions, ECS/Fargate, Timestream, Aurora/RDS, Amazon Bedrock, IoT Core, CloudWatch, and IAM
• Demonstrated data and ML engineering at scale, including distributed processing with Spark (PySpark) or equivalent, efficient handling of large time-series and binary signal data, and pragmatic reasoning about compute cost, runtime, and reprocessing risk
• Experience designing data mining and annotation workflows in partnership with domain experts, including label taxonomy design, labeling tooling, label quality control, and converting noisy expert feedback into model-ready ground truth
• Strong software engineering and programming skills in Python, with experience developing modular, testable, maintainable production code; proficiency with Git, code review, unit/integration/regression testing, debugging, CI/CD, and reproducible development practices on noisy, large, imperfect industrial datasets
• Proficiency in scientific and ML/DL frameworks and libraries such as NumPy, Pandas, SciPy, scikit-learn, TensorFlow, PyTorch, or similar tools
• Working knowledge of SQL (MySQL/Aurora, Athena) and the ability to independently explore, validate, and troubleshoot large and imperfect industrial datasets
• Excellent communication and teamwork skills, including the ability to collaborate with condition-monitoring domain experts, MLOps engineers, Data Engineers, Product, and other stakeholders
• Experience using Large Language Models, AI-assisted coding, and agentic AI tools for software and ML development
• familiarity with designing or integrating LLM- or agent-based workflows for data processing, analysis, decision support, or automation is highly desirable
• Experience building representative benchmark datasets, validating models against reviewed ground truth, analyzing false positives and false negatives, and creating repeatable evaluation or regression workflows is highly desirable
• Physics-informed ML and experience with additional modalities (electrical/current signature analysis, ultrasound, oil analysis) are a plus
• Demonstrated research work in high-quality journals or significant applied work on related subjects is a plus