In lithium-ion battery manufacturing, one of the most energy- and time-consuming steps is capacity grading: determining the capacity of each battery cell through full charge and discharge cycles. While this process ensures product consistency, it also requires expensive equipment and consumes a large amount of energy.
To address these challenges, a battery manufacturer developed an intelligent system to predict battery cell capacity using data generated during the earlier formation stage. By applying machine learning models trained on high-quality formation data, this solution can accurately forecast cell performance, eliminating the need for traditional grading.
This innovation significantly reshapes the company’s battery production line as follows:
-
Up to 90% of grading equipment can be removed
-
Energy consumption for grading drops by 90%
-
Production speed increases
-
Equipment investment, energy use, and maintenance costs are all reduced
Why TDengine
At the core of this solution lies a need to ingest and analyze large volumes of time-series data in real time. The team originally used MariaDB for data storage, but as data volumes scaled, performance degraded. After testing alternatives, they chose TDengine — a purpose-built time-series database — for its high ingestion speed, lightweight architecture, and advanced query engine.
TDengine enabled the system to:
-
Store one table per channel for granular data access
-
Use supertables for easy aggregation and batch comparison
-
Process real-time data streams for anomaly detection and early warnings
-
Easily visualize capacity distributions with real-time histograms
-
Integrate seamlessly with a Rust-based backend using TDengine’s Rust client library
Custom Algorithms + TDengine = No Need for Vector Databases
To ensure prediction accuracy, the system uses continuous machine learning (CML) to stay aligned with real-time production conditions. Since only a fraction of grading equipment is typically deployed, early-stage data is limited. A custom curve-matching algorithm solves this by retrieving similar historical data to predict cell capacity—enabling high efficiency with less infrastructure.
While this mirrors functionality found in vector databases, adding one would have increased system complexity. Instead, TDengine’s built-in aggregation, stream processing, and support for user-defined functions enable fast, efficient similarity searches without external tools.
TDengine also supports standard SQL and relational queries, making it easy to link analytical results with metadata. Its Rust connector ensures high-performance data writes. Altogether, TDengine delivers the performance needed—without the overhead of a separate vector database.
Real-Time Alerts and Efficient Visualization with TDengine
Battery production lines often lack real-time feedback, forcing supervisors to rely on manual reports every few hours to detect anomalies. This approach delays issue resolution and adds to operational overhead.
By using TDengine’s stream processing, key metrics like voltage and temperature can be analyzed the moment data is written—enabling automated alerts and timely detection. This not only reduces manual effort but also smooths out compute loads over time.
In addition, production teams closely monitor the distribution of predicted capacity values. As output scales, fetching and aggregating large volumes of data at the application layer becomes inefficient. TDengine solves this with built-in window queries and aggregation functions, allowing histograms to be generated directly within the database and returned quickly—ensuring fast, responsive visualization at scale.
As an example, the company used this code to generate a histogram:
let capacity_histogram_data: Vec<CapacityIntervalHistogram> = taos
.query(format!(
"SELECT histogram(capacity,'linear_bin','{}',0) as capacity_rang
FROM inference.`{}` WHERE ts > (NOW - {}d) ;",
serde_json::json!({
"start": min_capacity as i32,
"width": (max_capacity as i32 / total_capacity_intervals),
"count": total_capacity_intervals,
"infinity": false
}),
table_name,
ts_n_days
))
.await?
.deserialize()
.try_collect()
.await?;The output data was visualized on their frontend system:
Massive Cost Savings
By replacing traditional grading with a predictive system powered by TDengine, the manufacturing process saw a dramatic reduction in capital and operational expenses:
-
Equipment investment dropped from $7.44 million to $744,000
-
Annual energy costs were slashed from $1.09 million to $10,900
-
First-year savings exceeded $7.7 million
-
Cumulative second-year savings are projected to surpass $8.26 million
A Smarter Path Forward
Beyond just predicting battery capacity, the system enhances quality control and production monitoring. With TDengine, the team can detect anomalies in real time, reduce defects, and ensure long-term consistency in cell performance.
The success of this approach demonstrates how AI-powered analytics, combined with an efficient time-series database like TDengine, can transform even the most energy-intensive industrial processes, offering both technical and economic advantages at scale.
