AI & Advanced Analytics
AI on the Casting Floor: Shot-Profile Curve Anomaly Detection
AI Column
Instead of inspecting castings after they solidify, advanced foundries are deploying neural networks to analyze plunger speed-pressure profiles during the shot. By matching millisecond-level curve shifts to CT-scan histories, AI can predict structural porosity and die wear before the gate freezes.
In high-pressure die casting, quality assurance has traditionally been retrospective. The shot is poured, the part is ejected, it undergoes cooling, gate trimming, shot blasting, and machining. Hours or days later, a CT scan or leak test reveals a cluster of entrapped gas porosity. The part is scrapped, and the caster must account for the wasted energy, machining time, and alloy scrap value.
AI is shifting the quality gate from the inspection lab to the injection sleeve. By installing high-frequency sensors on the plunger cylinder and running real-time neural network inference, we can analyze the physics of the pour as it happens.
Visualizing the Injection Shot Profile Curve
Molten aluminum is injected into the cavity in under 100 milliseconds. Sensors capture position, piston velocity, and hydraulic pressure at a 2.5 kHz sampling rate. The chart below illustrates a healthy shot curve compared to an anomalous, porosity-forming curve.
Shot Cylinder Velocity Profile (Healthy vs. Anomalous)
An early drop in plunger velocity during Phase 1 (slow shot) indicates that a wave has formed in the cylinder, trapping air pockets that are subsequently injected into the casting cavity.
How Neural Networks Detect Floor Anomalies
A standard machine controller can only trigger alarms when values cross simple upper or lower thresholds. Neural networks look at the *relationship* between speed and pressure over time, isolating anomalies that simple sensors miss:
Premature Solidification (Cold Shuts)
If the molten alloy temperature is too low or die cooling is excessive, metal begins to solidify in the runner. The AI detects this by identifying a premature slope change in the pressure-displacement curve before the plunger reaches the transition point. It alerts the operator to adjust nozzle heat before the next cycle.
Turbulence-Induced Gas Porosity
If the plunger accelerates into the fast-shot phase too early, it creates a wave in the shot sleeve that traps air. The neural network detects this by identifying micro-oscillations in the plunger speed curve during the fast-shot fill phase, indicating the plunger is pushing through air pockets rather than solid liquid metal.
Intensification Phase Lag
To compact shrinkage porosity, the machine must apply high pressure immediately after the cavity fills. If the intensification pressure ramp-up is delayed by even 15 milliseconds due to hydraulic wear, the gate freezes and compaction fails. The AI flags this delay, identifying hydraulic cylinder slippage long before manual audits catch it.
Predicting Die Soldering with Thermal Imaging
Beyond plunger curves, AI models are also being connected to infrared cameras monitoring the die face after part ejection. Mold release spray wears off unevenly, leading to aluminum sticking to the steel die face (soldering).
The AI analyzes the thermal signature of the die face, identifying hot spots where heat has accumulated. If a hot spot remains above 320°C after ejection, the AI predicts soldering will occur within five cycles. It automatically signals the spray robot to apply localized coolant to that specific zone, preventing tool damage and saving hours of polishing downtime.
The ROI of Real-Time Casting Inference
Deploying AI to the casting floor does not require purchasing new presses. It involves retrofitting existing machines with high-frequency transducers and connecting the sensor outputs to an edge compute unit on the shop floor. By weeding out bad castings at the press and optimizing spray cycles, foundries report a 3% to 5% absolute increase in OEE—representing millions of dollars in saved capacity for large casting programs.