Understanding bolting curves: Typical patterns & their causes

Every bolted joint creates a curve. Torque over angle of rotation - a seemingly simple diagram that tells the complete story of a bolted joint. How hard is the material? Is there a head contact transition? Has the screw gripped? Does the nut rotate freely? Was there any jamming?

An experienced screwdriver technician reads this curve like an ECG. He can see immediately that something is wrong. But in most companies, the curve evaluation runs fully automatically according to simple limit values - target torque reached: IO. Otherwise: NOK. The actual information in the waveform remains unused.

This article explains eight typical waveforms that occur in series assembly: what they mean, what causes are behind them and what measures they must trigger. In addition: the complete diagnostic matrix, the IO/NIO evaluation scheme and an overview of common tightening methods with their characteristic waveform.

Basics: What a tightening curve shows

A tightening curve is the graphical representation of the torque (Y-axis) over the angle of rotation (X-axis) during a tightening process. It not only shows whether a connection has been tightened correctly - it also shows how the tightening process went.

The four phases of an ideal tightening curve

• Phase 1 - Free-running phase: The screw turns freely before the thread fully engages. Torque is minimal. Angle of rotation increases without a significant increase in torque.
• Phase 2 - Joining phase: The screw is pulled into the thread, connecting parts come into contact. Slight increase in torque, no stable linear range yet.
• Phase 3 - Elastic range: Head support established, screw is tensioned. Linear steep increase in torque over the angle of rotation. The gradient in this area is characteristic of the joint stiffness.
• Phase 4 - Completion: Target torque reached, screwdriver stops or switches off. For angle-controlled methods: defined angle of rotation turned according to pre-tightening torque.

The tightening processes and their characteristic curve shapes

The tightening method determines which waveform is considered 'normal' and which parameters are monitored. Different methods have fundamentally different curve characteristics.

8 typical curve shapes: Patterns, causes and measures

The following eight curve shapes cover the majority of all tightening problems that occur in series assembly. Each card shows the schematic curve profile (torque over angle of rotation), the typical causes and the recommended action.

Diagnostic matrix: Symptom → Cause → Verification → Immediate action

The diagnostic matrix is the tool for quickly finding the cause of tightening problems during production. From visible curve symptom to verifiable cause in under 10 minutes.

IO/NIO evaluation scheme according to VDI/VDE 2862

The evaluation of a screw connection according to IO (OK) or NIO (not OK) is carried out in accordance with VDI/VDE 2862 on the basis of several criteria simultaneously. Torque alone is not sufficient.

The torque indicates whether the screw is tightened. The curve indicates whether the connection is good. These are two very different questions.

-Amadeus Lederle CTE, CSP Intelligence GmbH

Before/after comparison: What process optimization makes visible in the curve

The value of a good curve evaluation is particularly evident in the before/after comparison following process optimization. Three real cases from practice illustrate how curve changes demonstrate process improvements.

Curve evaluation in practice: Manual vs. automatic vs. AI

Curve evaluation can take place at three levels - each has its place in production practice.

Level 1: Limit value-based evaluation (standard)

The most common method: The screwdriver evaluates IO/NIO exclusively according to limit values for torque and angle of rotation. Fast, robust and sufficient for low class requirements. Weak point: Curve shape is not evaluated.

A plateau in the rise, which indicates setting, is not recognized if the final torque is correct.

Level 2: Manual pattern recognition by experts

A trained technician checks conspicuous curves manually and recognizes patterns. Effective for root cause analysis of recurring NOK patterns. Not scalable for 100% curve analysis in ongoing production.

Typical: manual review sample or escalation for frequent NOK cases.

Level 3: AI-supported automatic pattern recognition

Machine learning models recognize curve deviations automatically - even beyond simple limit value checks. Training on historical curves with known evaluation. Also recognizes subtle patterns that escape human evaluators.

Prerequisite: sufficient amount of high-quality historical curve data (typically 10,000+ curves for stable model).

The hybrid recommendation for practice: automatic limit value evaluation for 100% of curves (level 1), manual expert review for conspicuous patterns and trend analyses (level 2), AI support for systematic pattern recognition for high volume and class A screw cases (level 3).

PRACTICAL TIP

IPM - Complete screwdriving curve infrastructure

CSP IPM records screwdriving curves completely and in real time: torque-rotation angle curve per tightening, per screw position, per serial number. Curves are stored unalterably, with tool ID, time stamp and IO/NIO evaluation - in accordance with VDI/VDE 2862 Class A/B/C.

• Real-time curve recording: complete progression per tightening, not just end values
• Curve comparison: current curve against reference curve per tightening position at a glance
• Trend analysis: change in pitch over production as an early warning for tool wear
• Cpk automatic: from torque and rotation angle values per tightening position
• Link to serial number: each curve is assigned to the unit produced - complete traceability
• AI integration possible: curve data as a training and inference basis for pattern recognition

→ Arrange a demo

Frequently asked questions

What is the difference between torque and preload force?

Torque (M_A, in Nm) is the mechanical force with which a screw is turned. It is the control variable of the tightening process. Preload force (F_V, in kN) is the actual tensile force built up in the bolt - what holds the connection together. Both are connected via the bolt equation: M_A = F_V × d/2 × (tan(φ + ρ') + ...), where φ is the pitch angle and ρ' is the friction angle. The problem: Friction varies. The same tightening torques therefore generate different preload forces depending on the lubrication, surface and material.

Why does the stick-slip effect cause the curve to become unsteady?

Stick-slip occurs when the static friction between the screw flanks and the thread is temporarily higher than the sliding friction. The connection 'sticks' (stick), builds up tension until the adhesion limit is exceeded, then slides jerkily (slip), reduces the tension again and the cycle starts again. In the curve diagram, this looks like high-frequency trembling. This is particularly common in stainless steel connections without lubricant because stainless steel tends to gall.

Can a connection be retightened - and does this change the interpretation of the curve?

Retightening is technically permissible for some connections, but it must be clearly marked as retightening in the protocol. A retightening curve looks different from an initial tightening curve: the freewheel area is minimal or non-existent, the rise begins immediately. IO/NIO criteria for retightening must be defined separately. VDI/VDE 2862 Class A: Retightening must be explicitly permitted in the work instructions and recorded separately. In practice, the following applies: If retightening is required frequently, there is a process problem.

How do I recognize galling (cold welding) in the bolting curve?

Galling appears in the curve as a strong, irregular rise often combined with a sudden drop (screw spins) or a very high pitch with jerking movements. It is an extreme form of the stick-slip effect, in which micro-welding occurs between thread flanks. Galling occurs particularly in stainless steel/stainless steel, titanium or aluminum/steel without sufficient lubrication. Prevention: suitable lubricant (MoS₂, PTFE), coated screws or change of material.

What does 'gradient' of the screwing curve mean - and how is it calculated?

The gradient (or slope) of the screwing curve is the rate of change of the torque per degree of rotation angle in the linear range of the curve: G = ΔM / Δα (Nm/degree). The gradient is a measure of the connection stiffness: stiff connection partner = high gradient, soft connection partner = flat gradient. A change in the gradient compared to the reference curve indicates changes in the connection geometry or the material properties - e.g. missing component, incorrect washer or changed material.

How many curves need to be saved and for how long?

According to VDI/VDE 2862 and IATF 16949 (class A): all curves per unit, unchangeable, for the product lifetime + legal retention periods (typically 15 years in the automotive sector). In file format: typically proprietary screwdriver formats (Bosch/Atlas Copco format), CSV with time series or standardized Q-DAS format. Data volume: one curve with 1000 measuring points at 1 ms sampling rate = approx. 8 KB. With 100 class A bolted layer suits per vehicle and 500 vehicles/day: ~400 MB/day. Over 15 years: approx. 2.2 TB - manageable, but archiving infrastructure required.

Can AI 'learn' incorrect assessments from historical curves and become worse as a result?

Yes - this is a real risk with AI-based curve evaluations. If historical curves that were evaluated as 'IO' actually had quality problems (but were not recognized at the time), the AI model learns this incorrect evaluation as a reference. The result: The model reproduces historical misjudgments. Countermeasures: Have training data validated by experts, include curves with known damage (e.g. field complaint attributed to screwdriver) as NOK class, check model performance regularly with validated test data.