ChronicleCare: Your Asset's Memory & Oracle
A personal, AI-guided conversational journal for equipment maintenance, helping individuals log sensory observations and historical data to predict asset health issues before they become critical.
Frankenstein's Predictive Pump
A low-cost predictive maintenance system for water pumps utilizing time-series data anomalies and a forgetting curve algorithm, designed to prevent catastrophic failures based on fragmented usage history.
Inspired by Frankenstein's piecing together of disparate parts and Memento's fragmented memory, this project focuses on predicting water pump failures using incomplete and sometimes unreliable historical usage data. Imagine a building manager grappling with inconsistent water pump maintenance records – dates are missing, repairs are poorly documented, and sensor data is intermittent. This project aims to stitch together this 'frankensteined' data to predict failures before they occur.
The concept revolves around a Raspberry Pi-based system connected to a water pump, gathering data like vibration, temperature, and pressure (using low-cost sensors). An 'Urban Traffic Data' scraper analogy is used to collect and clean relevant publicly available data, such as weather patterns (temperature, humidity), which can significantly impact pump stress. The 'Memento' aspect comes in with a custom-built forgetting curve algorithm. Instead of simply averaging historical data, the system assigns exponentially decaying weights to past data points, prioritizing recent information while retaining the 'memory' of long-term trends. More recent data is weighted more. However, the decaying weight prevents old data from overly skewing the prediction if the pump has since undergone maintenance or experienced significant changes.
Here's how it works:
1. Data Acquisition: Collect sensor data (vibration, temperature, pressure) from the water pump using low-cost sensors and a Raspberry Pi.
2. Data Cleaning & Enrichment: Use publicly available weather data (scraped from online sources) and manual log entries (pump usage hours, known repairs) to enrich the dataset.
3. Anomaly Detection: Implement anomaly detection algorithms (e.g., ARIMA, Exponential Smoothing, or a simple moving average with standard deviation bands) to identify deviations from the expected behavior. The 'frankensteined' nature of the data means the anomaly detection needs to be robust to missing values and outliers.
4. Forgetting Curve Prediction: Develop a custom forgetting curve algorithm. This algorithm assigns weights to past data points based on their age, recent repairs and manual logs. An event like a repair could reset the relevant parts of the pump 's memory. The weights are exponentially decayed as data ages, but manual log inputs can alter that curve. The prediction algorithm uses the weighted average of past data and anomaly scores to predict future pump failure probability.
5. Alerting: Based on the predicted failure probability, the system sends alerts to the building manager via email or SMS.
The niche here is focusing on systems with inconsistent or limited historical data – a common problem in older buildings. The low-cost implementation makes it accessible to smaller businesses or individuals. The high earning potential lies in offering this as a subscription service to building managers, promising to reduce maintenance costs and prevent costly emergency repairs. Imagine advertising: "Don't let your water pump become a monster! Preventative maintenance with Frankenstein's Pump!"
The concept revolves around a Raspberry Pi-based system connected to a water pump, gathering data like vibration, temperature, and pressure (using low-cost sensors). An 'Urban Traffic Data' scraper analogy is used to collect and clean relevant publicly available data, such as weather patterns (temperature, humidity), which can significantly impact pump stress. The 'Memento' aspect comes in with a custom-built forgetting curve algorithm. Instead of simply averaging historical data, the system assigns exponentially decaying weights to past data points, prioritizing recent information while retaining the 'memory' of long-term trends. More recent data is weighted more. However, the decaying weight prevents old data from overly skewing the prediction if the pump has since undergone maintenance or experienced significant changes.
Here's how it works:
1. Data Acquisition: Collect sensor data (vibration, temperature, pressure) from the water pump using low-cost sensors and a Raspberry Pi.
2. Data Cleaning & Enrichment: Use publicly available weather data (scraped from online sources) and manual log entries (pump usage hours, known repairs) to enrich the dataset.
3. Anomaly Detection: Implement anomaly detection algorithms (e.g., ARIMA, Exponential Smoothing, or a simple moving average with standard deviation bands) to identify deviations from the expected behavior. The 'frankensteined' nature of the data means the anomaly detection needs to be robust to missing values and outliers.
4. Forgetting Curve Prediction: Develop a custom forgetting curve algorithm. This algorithm assigns weights to past data points based on their age, recent repairs and manual logs. An event like a repair could reset the relevant parts of the pump 's memory. The weights are exponentially decayed as data ages, but manual log inputs can alter that curve. The prediction algorithm uses the weighted average of past data and anomaly scores to predict future pump failure probability.
5. Alerting: Based on the predicted failure probability, the system sends alerts to the building manager via email or SMS.
The niche here is focusing on systems with inconsistent or limited historical data – a common problem in older buildings. The low-cost implementation makes it accessible to smaller businesses or individuals. The high earning potential lies in offering this as a subscription service to building managers, promising to reduce maintenance costs and prevent costly emergency repairs. Imagine advertising: "Don't let your water pump become a monster! Preventative maintenance with Frankenstein's Pump!"